Luminous Landscape Forum

Equipment & Techniques => Cameras, Lenses and Shooting gear => Topic started by: AlfSollund on January 21, 2012, 01:01:10 pm

Title: Does a photo give spatial information (the nose job)?
Post by: AlfSollund on January 21, 2012, 01:01:10 pm
Here are questions based on a lengthy perspective discussion. Please feel free to ignore if you like:

1) Does a photo give any spatial information? That is, can we from a normal 2-D photo get any kind of information about distances and objects spatial placement?
2) Does a wide give different or more spatial information than a tele?

And here are answers:
1) No, non whatsoever. The information can be used to guess about spatial, but no information can be derived for sure. Its no way of knowing the relative spatial positions to objects in a photo, or even if they are spatial.
2) No, since the answer on 1) is zero, the answer on 2 will also be zero. So by zooming you are not deriving any more spatial information, you are simply adding different information.

Here i an easy way to prove this. Look at any given photo. Based on the photo itself its no way of saying that this is a photo of a photo where all objects are in the same plane, or a photo of real 3-D objects. I can zoom all I like while taking the photo, but the difference will add no clues as to I'm seeing a photo of a photo or not.

This is the reason that we and the rangefinder cameraes have 2 eyes. These gives different perspectives and add the needed information to see spatially. Actually much of what we are doing when creating a photo is trying to make the illusion of spatial information by having varnishing lines etc in the photo.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: bill t. on January 21, 2012, 02:45:32 pm
I will ignore fact that I can look at an image any number of generations removed from the original, and still ferret out the relative depth of objects within the scene.  Any ordinary human visual cortex or even just your average biggest supercomputer in the world can do that.  Ok, very subjective, and the objects in the image have to be contained within the set of things recognizable to humans and/or super-computers.  So all bets off for abstractions.

But, quantification can be had by meticulously analyzing texture and edge contrasts within the scene, thereby inferring relative amounts of defocus for various objects, which tells us something about depth.  We need know nothing for sure about the image or its history, it's all relative based on a self-benchmarking source.  Less focused objects are either closer or farther away than more focused ones, and simple trending analysis can be used to infer which direction.  With enough hardware and computing power, it's a piece of cake.  Hah!   :)

Last but not least, there is a lot of proprietary software out there that can do this stuff on digitally scanned images, mostly developed by wiz kids to convert single camera motion pictures into (admittedly awful) approximations of what should have been shot as two-camera 3D in the first place.  The algorithms benefit greatly from facial recognition, etc.  Computers do most of the work, with only a little help from organic, carbon-based units.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: jerryrock on January 21, 2012, 03:07:45 pm
I think this camera does give spacial information.

https://www.lytro.com/science_inside
Title: Re: Does a photo give spatial information (the nose job)?
Post by: mouse on January 21, 2012, 07:48:57 pm
1) Does a photo give any spatial information? That is, can we from a normal 2-D photo get any kind of information about distances and objects spatial placement?
2) Does a wide give different or more spatial information than a tele?


Can you tell us exactly how you define "spatial information"?
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Wayne Fox on January 21, 2012, 08:33:31 pm
well, it seems fairly accurate spacial information could be derived from in image if you know the actual dimensional details of the items in the photograph.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: John Camp on January 22, 2012, 12:36:33 am
Yes. Check this:

http://en.wikipedia.org/wiki/Photogrammetry
Title: Re: Does a photo give spatial information (the nose job)?
Post by: AlfSollund on January 22, 2012, 03:16:14 am
Can you tell us exactly how you define "spatial information"?

3-D information about the objects in the photo, such as the distance to the objects and their relative 3-D position to each other and the photographer (even such a trivial information if one object is in front or behind another object, where the object also might be the photographer). Not included in my definition: That objects must be more than zero distance from the photographer.

well, it seems fairly accurate spacial information could be derived from in image if you know the actual dimensional details of the items in the photograph.
So how will we do this if I take a photo of a photo? There is only one object in this photo, and that is a (near) 2-D object. namely the photo i photograph. Secondly there are methods to guess the actual size, but not possible to know for sure.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: mouse on January 22, 2012, 05:31:06 am
3-D information about the objects in the photo, such as the distance to the objects and their relative 3-D position to each other and the photographer (even such a trivial information if one object is in front or behind another object,

Emphasis mine.  The easiest and most obvious contradiction to your argument is when one object is partially obscured by a closer object.  Surely this indicates the relative 3-D position.  Granted one must have additional information to determine exact distances between objects, but your original argument concerns relative distances; i.e. is one object closer to the camera than another? 

You did ask "Does a photograph give any spatial information"?
That's why I asked for your definition of spatial information.  Relative information is still information.  Bigger is information even if you can't say how much bigger.

And your example of a photograph of a photograph is a red herring.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: BJL on January 22, 2012, 08:40:19 am
In practice even a one-eyed view is often enough to draw some fairly reliable conclusions about subject distances. The obvious case is knowing the relative sizes of various objects in the scene (the famous nose-to-ear ratio), but out-of-focus effects are common visual clues too, and I would think that measurements of blurring at edges could reveal something. To be pedantic, a given degree of OOF blurring is seen with objects at two different distances, in front of and behind the plane of exact focus, but combined with other information like which objects blocks the view of which others, something can be learnt ...

... though not with the Degree of certainty required to get someone to admit that they were wrong in a long forum discussion.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: OldRoy on January 22, 2012, 10:06:43 am
I know better than attempt to participate in the kind of exchange that frequently occurs on LL; the adjoining focal length/perspective discussion is a case in point. However nostalgia overcomes my reluctance.

Someone posted a link about photogrammetry. Whilst by no stretch of the imagination am I an expert in this field I was involved, many years ago (and in retrospect entirely prematurely) in a project to implement stereo photogrammetry using TV technology. This was at a time when the technology was predominantly analogue with the exception of a few monochrome CDD cameras.

Some of my colleagues were senior academics with a deep understanding of the mathematics and optics involved (projected geometry). Our objective, roughly speaking, was to derive real time (or nearly so) 3d measurements from a pair, or more, of TV cameras. This is analogous to the techniques employed for a very long time in the creation of 3d contour maps where pairs of aerial photographs, taken by a camera of known optical characteristics, are combined in a stereo viewer and light spots aimed at each image are converged. From the x/y data related to the location of the reference point in each image, a z value may be calculated. Assuming known geometry of the cameras/images, the equivalent x/y/z values for the terrain can be derived.

I put a lot of time and money into this project. It was pretty obvious that the same technique, if it could generate values quickly, could be used in a wide variety of applications. In manufacturing, say, engine blocks, co-ordinate measuring machines are used to "poke" at key locations on the target and the deflections compared to a reference set of measurements. Unfortunately this sort of method (which I believe is still used) is subject to "drift" and a requirement for recalibration. My own proposal was to use a raster scanned laser to mark the targets. We also had a lot of interest in using this method for surveying the submerged portion of oil platforms. At one point a friend's film effects company that was involved had a visit from someone at NASA to take a look at a stereo TV viewing system we had attached to a manually operated robotic arm. Unfortunately in the mid 1980s the limitations of the technology were a big, if not entirely un-surmountable obstacle.

I know that this technology has come a long way but I'm always surprised how long it has taken for similar techniques to come into widespread use. At the time I was involved some of my contacts were mainly interested in 3dTV for entertainment use. Of course now most of the hardware required is available off the shelf and the requisite computing power is no longer even a consideration...

Roy
Title: Re: Does a photo give spatial information (the nose job)?
Post by: AlfSollund on January 22, 2012, 11:02:44 am
Emphasis mine.  The easiest and most obvious contradiction to your argument is when one object is partially obscured by a closer object.  Surely this indicates the relative 3-D position.  Granted one must have additional information to determine exact distances between objects, but your original argument concerns relative distances; i.e. is one object closer to the camera than another? 
Thats the point. From one image only you cannot see if a object is obscured  by another, or closer to the camera. You can guess, and have an indication. Its quite easy to make objects appear to be obstructed. You cannot even tell if the object is in front of the photographer. By front I mean the way the lens and nose in pointing when taking the pic.

You did ask "Does a photograph give any spatial information"?
That's why I asked for your definition of spatial information.  Relative information is still information.  Bigger is information even if you can't say how much bigger.
Still no spatial information. Only information that can be used for guessing. You cannot say if one object is bigger than another from a photo. Even the appearance of a human cannot be trusted by 2-D only. You can only say if one object is larger than another in the 2-D representation.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ellis Vener on January 22, 2012, 11:16:16 am
Spatial information gleaned from a photograph is one of the pillars of both photo reconnaissance as well as military and coporate intelligence gathering. Light and shadow length are important.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Bart_van_der_Wolf on January 22, 2012, 12:07:02 pm
Spatial information gleaned from a photograph is one of the pillars of both photo reconnaissance as well as military and coporate intelligence gathering. Light and shadow length are important.

That's correct, as is the amount of (de-)focus:
http://spie.org/x648.html?product_id=736131 (http://spie.org/x648.html?product_id=736131) or
http://www.comp.nus.edu.sg/~zhuoshao/depthRecovery/index.html (http://www.comp.nus.edu.sg/~zhuoshao/depthRecovery/index.html).

Cheers,
Bart
Title: Does a photo give spatial information: not if you demand philosophical certainty
Post by: BJL on January 22, 2012, 01:20:15 pm
From one image only you cannot see if a object is obscured  by another, or closer to the camera. You can guess, and have an indication. ... Only information that can be used for guessing.
Almost everything that we claim to know is merely a "guess" by this standard of "not absolutely certain". Many philosophers would declare this to be the case for any claims of "a postiori", empirical knowledge about the physical world. Skeptically speaking, empirical science is based entirely on "information that can be used for guessing". This is why uncomfortable scientific conclusions about the health effects of cigarette smoking or the evolutionary history of life on earth can be dismissed as mere guesses/conjectures/hypotheses if one so desires.

For example, even the spatial relationships that we think we detect with stereoscopic vision can in fact be illusions (as with glasses-free 3D TV presentation of a computer generated 3D movie), but most people would not then conclude that photogrammetry is pure guesswork.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: mouse on January 22, 2012, 06:17:39 pm
From one image only you cannot see if a object is obscured  by another, or closer to the camera. .... Even the appearance of a human cannot be trusted by 2-D only. You can only say if one object is larger than another in the 2-D representation.

Are you really contending that, in a photo, if one object partially obscures another, one cannot conclude that the partially obscured object is further from the camera?  Imagine for example, a photograph of a person standing in front of, and partially obscuring the lower part of a tree.  Viewing only the photo I suppose one could reach alternate conclusions:  The person is in front of the tree (closer to the camera) OR the tree is growing out of the persons head.  So granted, a photo cannot give any information about spatial relations in the real world if the observer has no experience with the real world.  In what galaxy do you reside?

You are, I assume, the same AlfSollund who posted several times in a contemporary thread on the subject of perspective.  In that thread you were in firm agreement with the majority on the meaning of perspective and the fact that it depended on distance from observer to subject, and not on the focal length of the lens.  You posted:

Quote
Mouse is 100% correct. Why anyone would spend any efforts in trying to contradict facts is beyond me.

The difference between two photos with different focal lengths is that tele one has a subset of of the content, with exactly equal composition and perspective as the wide (DoF might differ). And nobody else than the ones ignorant of perspective would want to compare these, since the result is given as most of us know.

I'm sorry to say that you are arguing with the last 1000 years or so of knowledge on simple geometry 

And with this Im do not want to give any more education lessons of subjects that should be obvious at ground school levels, and that the Greeks mastered more than 2000 years ago (and probably other cultures before that).

I assume you are also in agreement with the common definition of perspective; that it involves visual distortions in a 2D image which provide information about the relative spatial positioning of objects in the original 3D scene.  Here again it is critical that the observer is cognizant of the significance of such visual distortions based on what he has seen and experienced in the real 3D world.

Now, in a separate thread, you are essentially arguing that perspective does not exist.  To paraphrase, you are arguing that "a photo cannot give any spatial information.  That we cannot derive from a photo any kind of spatial information about distances and objects spatial placement."

From this abrupt turnaround I can only conclude that you are badly confused, are unable to understand the meaning of the words (yours and others) that are being used or, perish the thought, you are simply trolling. :'(
Title: Re: Does a photo give spatial information (the nose job)?
Post by: j-land on January 22, 2012, 09:29:41 pm
This (http://ahsmail.uwaterloo.ca/kin356/illusion/Ames.HTML) comes to mind...
Title: Re: Does a photo give spatial information (the nose job)?
Post by: tom b on January 22, 2012, 09:32:57 pm
Check out this site (http://make3d.cs.cornell.edu/)!

Cheers,
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ray on January 23, 2012, 09:00:22 pm
I'm surprised that no-one seems to have mentioned the obvious fact that a 'standard lens' provides the closest representation of perspective and distances for the human eye. That's probably why it's called a 'standard lens'.

Of course, a 'standard lens' can be almost any focal length depending on the format of the camera. On a P&S camera it might be 5mm. On the so-called cropped format, around 30mm; on full-frame 35mm format, around 50mm; on Medium Format around 80mm; on 4"x5" format around 150mm; and  on 8"x10" format, around 320mm.

In other words, one camera's standard lens could be another camera's wide-angle or telephoto, within the limitation of the lens image circle for wide angles, and other design factors.

Without the use of computer analysis and the aid of other measuring devices, a human being, using his own eyes, is able to guess only fairly accurately the relative distances of objects provided he is familiar with at least some of the objects in the scene.

When people who have been born blind as a result of some defect which can later be corrected through modern surgery, one might think what a wonderful blessing that would be. Unfortunately, the reality is that what such people see with their gifted sight is a meaningless jumble which can be very disturbing. Apparently, it's a slow and arduous process of relearning that such people have to go through, as a baby does as it gradually works out that one object that is obscured by another is further away than the object that does the obscuring.

As I look out of my hotel window, a huge ceiling-to-floor window, and peruse the view of skyscrapers, trees, carports, backyards of smaller buildings, parked cars and occasional people walking in the near foreground below, I am able to guess reasonably accurately the distances between myself and any object in the scene. There are hundreds of clues, the most ubiquitous of which are the windows and balconies indicating the position of each floor in a building and the total number of floors. From experience I know that each floor is 3 to 4 metres in height.

The furthest skyscraper appears to my eyes to be 6 or 7 hundred metres away. Regardless of the accuracy of this estimate, I could produce the same estimate when viewing a photo of the scene in front of my eyes taken with a 'standard lens' from the same position.

If I'm presented a photo of the same scene shot from the same position, but taken with a wide-angle lens, it becomes much more difficult to estimate the distance of that furthest skyscraper. It really does look further away. Instead of 600 metres it could be 2 or 3 kilometres away, if taken with a really wide-angle lens. However, as a photographer, knowing that the shot was taken with a wide-angle lens, I can make allowances for that fact. I might think, although the skyscraper actually looks as though it's 2 or 3 kilometres away, in reality it's probably less than 1 kilometre.

Conversely, if the skyscraper is shot with a telephoto lens, whether an actual telephoto lens or an equivalent crop from a wide-angle shot, estimating the distance to the skyscraper is also problematical because it really does look much closer. Again, knowing that the skyscraper was shot with a telephoto lens, I can make allowances and claim that the skyscraper is really much further away than it appears.

This is why I claim that it is absurd to say that focal length of lens has no bearing on perspective as experienced by the human observer.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: mouse on January 23, 2012, 10:58:18 pm
I think I finally understand Ray's problem.  To appreciate it fully you need to read some of his posts in  this thread (http://www.luminous-landscape.com/forum/index.php?topic=61222.0)


While the majority of photographers, painters, and others involved in creating or interpreting 2D images have adopted the standard, mathematically defined concept of (linear) perspective, Ray has created his own personal definition of perspective..  He has every right to do so, but it is neither legitimate nor rational for him to argue that the consensus definition is wrong while his is correct.  While Ray may continue to submit any amount of evidence in support of his personal definition, nothing he has written calls into question the legitimacy of standard definition nor suggests that his definition is more useful or appropriate.  It is axiomatic that any discussion must remain fruitless if you cannot reach agreement on the basic meaning of the subject under discussion.

We encounter the same problem in this thread with AlfSollund's thesis that a photo cannot convey any spatial information.  I believe the very fact that he makes this statement is evidence that his definition of spatial information is quite different from the concept shared by the majority of photographers (and other observers).  However absent any precise definition of what Alf means by spatial information (I asked him, but have not yet received a clear answer), it becomes impossible to engage in any sort of a logical discussion.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ray on January 24, 2012, 05:01:53 am
While the majority of photographers, painters, and others involved in creating or interpreting 2D images have adopted the standard, mathematically defined concept of (linear) perspective, Ray has created his own personal definition of perspective..  He has every right to do so, but it is neither legitimate nor rational for him to argue that the consensus definition is wrong while his is correct.  

I assure you that there is no-one more interested in the possibility that I am wrong. If my eyesight and sense of perception are faulty, I'm the first one who wants to know. If my reasoning is faulty or illogical, or the evidence I base my reasoning upon is not factual, I'm the first one who wants to know.

I see no merit in maintaining an illogical and incorrect position on any matter just for the egotistical satisfaction of appearing to be right or winning the argument.

You will find many references on the internet expressing the opinion that the standard lens (45 or 50mm for 35mm format) provides the sense of perspective that most closely matches human vision. However, since I'm not the sort of person to blindly accept any opinion as being true simply because there is a consensus on the matter, I've tested this for myself, taking numerous shots of the same scene at different focal lengths, and found that 50mm (on 35mm format) does indeed more closely match what I see in reality, in terms of spatial relationships and in terms of assessing the real distance to the viewer (me) of the subject that's been photographed.

Now, it could be that I'm part of a minority group of people who are 'perspectively' challenged, and that most people are able to immediately see the true spatial relationships and distances in a photographic image, whatever focal length of lens has been used.

If this is the case, then in my defense I will mention that Leonardo da Vinci appears to have had a similar problem, according to the following extract from the University of Chicago. However, the telescope hadn't been invented in his time.

Quote
Leonardo da Vinci, writing soon after the invention of scientific perspective, dismissed it as perspectiva accidentalis, and in his work Trattora della Pittura noted the distortive effects of perspective in wide angles and the various visual manipulations and elisions that occur from arbitrarily moving the constructed vanishing point in a painting. Leonardo encouraged painters instead to focus on parallel developments in aerial perspective – gradations in color, shadow, and texture to denote three-dimensional relations.

It would be interesting if you could tell me which of the following statements, upon which I base my reasoning, is incorrect.

(1) The standard lens, 50mm for 35mm format, most closely matches the natural perspective of the human eye.

(2) Shots from wide-angle lenses, uncropped, make distant objects appear further away and make close objects appear closer than they actually are in reality.

(3) Cropping any image, whether in camera or in post-processing, is effectively no different than using a longer focal length of lens that provides the same angle of view as the cropping.

(4) The effective focal length of any lens is a relationship between its actual focal length, the size of the sensor, and/or the degree of cropping in post-processing.

(5) It's not the focal length of the lens per se that has any bearing on perspective, but the effective focal length.

It would help if you could specify which of the above statements is wrong in your opinion. We could then concentrate on the problem area instead of going round in circles.


Title: Re: Does a photo give spatial information (the nose job)?
Post by: fredjeang on January 24, 2012, 05:56:20 am
(1) The standard lens, 50mm for 35mm format, most closely matches the natural perspective of the human eye. Correct, but it is more a 70mm 35mm equivalent that matches closer the human eye. although it is admited the 50 as a reference.

(2) Shots from wide-angle lenses, uncropped, make distant objects appear further away and make close objects appear closer than they actually are in reality. in part correct: in fact the distance between the planes being bigger wich gives the sensation that what is close is closer.
If you measure with a ruler an object, let's say a canva and you take the picture in PS and choose a scale, when measure with the PS ruler, you'll have the same value as the real object you measured. In other words, dimensions are respected.
So a photo gives effectively a spacial information regardless of the lens used. Studdied carefully, it gives distances-dimensions informations that matches with the real world. The only thing you have to have is a reliable reference point: you have to know a real dimensions of at least one object in the scene. You can try that easily at home, shooting a room.

A software like Nuke is capable of creating a 3D space from a 2D image. It has to know the focal lenght used. It has a database of lens brands in wich it calculates the correspondant 3D space in wich the picture was shooted. Therefore, it has the capabilities to do stichings in 3D without 3D infos because the 2D are actually "containing the infos" as long as you know the lens used. Here is how it works: http://www.youtube.com/watch?v=vW7K-AwFAuM&feature=related

(3) Cropping any image, whether in camera or in post-processing, is effectively no different than using a longer focal length of lens that provides the same angle of view as the cropping. Correct, cropping in post is eactly the same as using a longuer focal at the exact same camera position. The only difference is that you'd have a smaller image dimension for the crop process.

(4) The effective focal length of any lens is a relationship between its actual focal length, the size of the sensor, and/or the degree of cropping in post-processing. Correct

(5) It's not the focal length of the lens per se that has any bearing on perspective, but the effective focal length. Don't really understand what you wrote here so can't opinate. The perspective has nothing to do with the lens and never has. It has all to do with the spacial point of view where the picture is taken from. If you don't move and you change your lenses, your perspective will be exactly the same. It only changes if you move in space, regardless of the lens.
That's what happens in tilt-shift as to correct-alter the perspective, a displacement has to occur, wether a physical displacement of the camera or the lens itself, wich is the case in view-tech cameras and tilt-shift lenses.
What yes changes when you change lens is the D.O.F but as long as you stay at the same place, the perspective is the same.(see your point 3)




Title: Re: Does a photo give spatial information (the nose job)?
Post by: hjulenissen on January 24, 2012, 08:15:54 am
We encounter the same problem in this thread with AlfSollund's thesis that a photo cannot convey any spatial information.  I believe the very fact that he makes this statement is evidence that his definition of spatial information is quite different from the concept shared by the majority of photographers (and other observers).  However absent any precise definition of what Alf means by spatial information (I asked him, but have not yet received a clear answer), it becomes impossible to engage in any sort of a logical discussion.
A camera-sensor is a 2-d array of sensels. In that sense, there is not enough information to specify general 3-dimensional structures with certainty.

3-d structures (generally) tend to be a lot less general than one might think. If one use this assumption, you may be able to make well-educated "guesses" as to what the scene really looks like. Especially if you are satisfied with a "2.5D" luminance + distance estimate. A good example of when this breaks down is if you take a snap of a painting or a photograph. Every part of the in-camera jpeg file may suggest that it is a castle in southern France, but the true shape of the object is flat...

Using focus information may help the procedure. I believe that one of the focus stacking software packages can do faux 3-d models from a series of stacekd focus images.

-h
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ray on January 24, 2012, 09:38:21 am

(2) Shots from wide-angle lenses, uncropped, make distant objects appear further away and make close objects appear closer than they actually are in reality. in part correct: in fact the distance between the planes being bigger wich gives the sensation that what is close is closer.
If you measure with a ruler an object, let's say a canva and you take the picture in PS and choose a scale, when measure with the PS ruler, you'll have the same value as the real object you measured. In other words, dimensions are respected.
So a photo gives effectively a spacial information regardless of the lens used. Studdied carefully, it gives distances-dimensions informations that matches with the real world. The only thing you have to have is a reliable reference point: you have to know a real dimensions of at least one object in the scene. You can try that easily at home, shooting a room.


Not sure I follow. Let's say I take a photo of a skyscraper that in reality is 700 metres away but appears to be about 2 kms away because I used a 14mm lens (on 35mm format). If there were a sufficient number of clues, that is objects of known dimension in the image, I can see how it would be possible to calculate all distances using a ruler. For example, if the image resolution were sufficient to enable me to count the number of floors in the skyscraper, I could determine the height of the skyscraper, approximately, provided I could see all the floors.

If there was a straight road leading from where I took the shot, filled with cars in a traffic jam, leading all the way to the base of the skyscraper, I could count the cars, if the resolution were sufficient, and provided the elevation of my perspective allowed me to see all the cars. I could estimate an average length for each car including the space between them, and get a rough estimate of the true distance of the skyscraper.

But these are all big 'ifs'. Supposing only part of the skyscraper were visible above a line of trees in the foreground which completely obscures everything behind the trees, apart from the tops of tall buildings?

In any case I'm not arguing that such relative sizes and distances could not be calculated if sufficient information were available, including I presume the focal length of the lens and type of camera, or lens equivalent.

Quote
A software like Nuke is capable of creating a 3D space from a 2D image. It has to know the focal lenght used. It has a database of lens brands in wich it calculates the correspondant 3D space in wich the picture was shooted. Therefore, it has the capabilities to do stichings in 3D without 3D infos because the 2D are actually "containing the infos" as long as you know the lens used. Here is how it works: http://www.youtube.com/watch?v=vW7K-AwFAuM&feature=related

The key point here is that lens and camera information is required. This software does not appear to work on the basis that perspective is unrelated to lens focal length.

Quote
The perspective has nothing to do with the lens and never has. It has all to do with the spacial point of view where the picture is taken from. If you don't move and you change your lenses, your perspective will be exactly the same. It only changes if you move in space, regardless of the lens.

How do you reconcile your statement above with your answer to my first question, ie.
Quote
Correct, but it is more a 70mm 35mm equivalent that matches closer the human eye. although it is admited the 50 as a reference.

Let's not quibble about whether it's closer to 70mm or 50mm. The point I would make is, if there is a particular focal length of lens (whatever it may be precisely) that matches the perspective as experienced by the human eye, then all focal lengths of lenses which are different to that will not match the natural perspective of the human eye. Therefore focal length of lens has a bearing on perspective, as experienced in a human observer.

Title: Re: Does a photo give spatial information (the nose job)?
Post by: fredjeang on January 24, 2012, 10:31:30 am
Because:

The wide angle doesn't affect the distances, as no other lens in fact. It gives the sensation that it does affect and therefore that the distances between close and far are bigger but indeed that's not what is happening. It just affect the scale, but proportionally distances are the same.

To verify that fact, you need to put your camera on a tripod and do not change the location. Start with a 50mm-70mm to have a reference closed to the human eyes. Take a picture.
Then, put a wide in your camera without moving nothing. What do you see? the perspective is exactly the same, exactly. But, it recorded a wider angle of view. Your 50mm shot is contained within the wide shot and if you put it on a layer in PS you'd see that the 50mm shot is excactly a part of the wide angle shot with zero perspective change. As your angle of view is bigger, but the sensor that records it has not changed, it's the same size, the objects are appearing further BUT it's an ilusion, if you measure the distances, you will obtain the same than with the 50mm, than with a 1000mm, and it will match the real world.

If you have a painting wich size is 50cm x 40, it's 50x40 regardless of the lenses.

You can actually do this testing at home.

Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ray on January 24, 2012, 11:15:34 am
Because:

The wide angle doesn't affect the distances, as no other lens in fact. It gives the sensation that it does affect and therefore that the distances between close and far are bigger but indeed that's not what is happening. It just affect the scale, but proportionally distances are the same.

To verify that fact, you need to put your camera on a tripod and do not change the location. Start with a 50mm-70mm to have a reference closed to the human eyes. Take a picture.
Then, put a wide in your camera without moving nothing. What do you see? the perspective is exactly the same, exactly. But, it recorded a wider angle of view. Your 50mm shot is contained within the wide shot and if you put it on a layer in PS you'd see that the 50mm shot is excactly a part of the wide angle shot with zero perspective change. As your angle of view is bigger, but the sensor that records it has not changed, it's the same size, the objects are appearing further BUT it's an ilusion, if you measure the distances, you will obtain the same than with the 50mm, than with a 1000mm, and it will match the real world.

If you have a painting wich size is 50cm x 40, it's 50x40 regardless of the lenses.

You can actually do this testing at home.




Fred,
Surely you don't believe there is anyone who thinks for one moment that changing lenses on a camera changes the actual distances between an observer and the subject. You'd have to be stark raving bonkers to believe that. Only a massive earthquake could do that, and then only marginally.

I'm talking about the appearance of perspective in a photographic image, taken from the same position with different focal lengths of lenses.

An increase in the FL of lens, which you've agreed is the same as cropping, results in a different perspective, as in the two images below. One is taken with a 10mm equivalent by stitching 3 images taken with a 14mm lens, camera vertical. The other is taken with a 50mm lens; both on the full frame D700.

Since I was there and took the shots, I can tell you with complete certainty that the perspective in the 10mm equivalent stitch is quite different from what I see from exactly the same position I used the camera, whereas the perspective and relative sizes of objects in the 50mm shot is a very close match to what I see when I look through the window of the 6th floor apartment from where I took the shots.

Now, I understand perfectly if I crop the 10mm shot to the same angle of view as the 50mm shot, I get the same effect, apart from resolution. And why shouldn't I? Cropping is the same as using an actual lens with the same angle of view as the crop. We've already agreed on that.

If I crop the 10mm image to the same FoV as the 50mm images, and find the perspective is the same, all I have proved is that two 50mm lenses used from the same position provide the same sense of perspective. That's so obvious it hardly needs confirming.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: fredjeang on January 24, 2012, 11:43:23 am
While you do not change your point of view, (means no pshysical displacement of any means) the perspective is the same for all focal lenses, wide, tele, whatever. To alter the perspective you need to displace your point of view.
Then you have distortions that occurs (different according to the lens-brand-price etc...) but they have nothing to do with the perspective itself as long as you stay at the same place.

So, if your point of view is the same, your perspective will always be constant between a 50 and a 200mm, a 200 and a 20.

It's not that 2 50mm at the same position provide the same perspective, it's that any focal lengh at the same position provide the same perspective.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Bart_van_der_Wolf on January 24, 2012, 11:54:03 am
In any case I'm not arguing that such relative sizes and distances could not be calculated if sufficient information were available, including I presume the focal length of the lens and type of camera, or lens equivalent.

All that's required is the focal length, as it determines the magnification factor. Magnification M = f/(s-f), where s is subject distance and f is focal length. That combined with an inkling of a clue of what we're looking at is enough for human vision to derive a sense of dimension, and hence perspective since only position determines perspective, but do read on.

What is probably confusing your concept of perspective, is that there is also something like projection distortion, which is what DaVinci and his contemporates were getting a grasp on. Unlike the human eye, a lens (or a pinhole) projects its image on a flat plane. That means that only when the output is looked at from the wrong viewing distance, things look distorted (the perspective of vanishing points seems distorted, e.g. a wide-angle or a tele 'look'). Since short focal lengths have their corner rays strike the flat plane at a much more oblique angle, that effect is even stronger (because we're usually looking at the result from 'too far away' to maintain the geometrical integrity). Objects may look stretched out, people look fat, noses look out of proportion.

When we look at the image on a flat plane from the proportionally scaled distance, the sense of perspective is identical for all focal lengths. Proportionally scaled is when we multiply the sensor size by a factor M, then we need to view it from a distance of focal length times M as well (which may get unpractical for ultra wide-angle and telelens shots).

The whole concept is formally called anamorphic projection distortion, or anamorphosis (http://en.wikipedia.org/wiki/Anamorphosis).

Cheers,
Bart
Title: Re: Does a photo give spatial information (the nose job)?
Post by: fredjeang on January 24, 2012, 12:20:38 pm
All that's required is the focal length, as it determines the magnification factor. Magnification M = f/(s-f), where s is subject distance and f is focal length. That combined with an inkling of a clue of what we're looking at is enough for human vision to derive a sense of dimension, and hence perspective since only position determines perspective, but do read on.

What is probably confusing your concept of perspective, is that there is also something like projection distortion, which is what DaVinci and his contemporates were getting a grasp on. Unlike the human eye, a lens (or a pinhole) projects its image on a flat plane. That means that only when the output is looked at from the wrong viewing distance, things look distorted (the perspective of vanishing points seems distorted, e.g. a wide-angle or a tele 'look'). Since short focal lengths have their corner rays strike the flat plane at a much more oblique angle, that effect is even stronger (because we're usually looking at the result from 'too far away' to maintain the geometrical integrity). Objects may look stretched out, people look fat, noses look out of proportion.

When we look at the image on a flat plane from the proportionally scaled distance, the sense of perspective is identical for all focal lengths. Proportionally scaled is when we multiply the sensor size by a factor M, then we need to view it from a distance of focal length times M as well (which may get unpractical for ultra wide-angle and telelens shots).

The whole concept is formally called anamorphic projection distortion, or anamorphosis (http://en.wikipedia.org/wiki/Anamorphosis).

Cheers,
Bart

Exactly, and well described.

When student, I used to work with this master: http://www.tjeerdalkema.net/ . He is a recognized specialist in anamorphosis. We where getting a little bit of money assisting him to mount the structures. Calculations where extremely complicated. He was calculating on the studio floor that was full of geometrical equations.

The confusion often comes from the distortions factors, but in fact the perspective is only affected by a change in the position.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ray on January 24, 2012, 01:22:44 pm
All that's required is the focal length, as it determines the magnification factor. Magnification M = f/(s-f), where s is subject distance and f is focal length. That combined with an inkling of a clue of what we're looking at is enough for human vision to derive a sense of dimension, and hence perspective since only position determines perspective, but do read on.

Bart,
That doesn't really make sense to me. Surely a 360mm lens on 8"x10" format has the same degree of magnification as a 50mm on 35mm format. You need to know more than the FL of lens, don't you?

Quote
What is probably confusing your concept of perspective, is that there is also something like projection distortion, which is what DaVinci and his contemporates were getting a grasp on. Unlike the human eye, a lens (or a pinhole) projects its image on a flat plane. That means that only when the output is looked at from the wrong viewing distance, things look distorted (the perspective of vanishing points seems distorted, e.g. a wide-angle or a tele 'look'). Since short focal lengths have their corner rays strike the flat plane at a much more oblique angle, that effect is even stronger (because we're usually looking at the result from 'too far away' to maintain the geometrical integrity). Objects may look stretched out, people look fat, noses look out of proportion.

I don't believe so. I'm already aware of such differences, which is why I quoted the following on another thread, which explains these differences quite well. The bold type has been added by me.

Quote
This discrepancy between camera and physical eye is accounted for in part by the fact that in our eyes, light projects not onto a flat surface, but the curved inner surface of our eyeballs. Furthermore, a large portion of our perception comes from having two eyes that can triangulate relative depth (known as stereopsis, which is a form of parallax), and the ability to move our heads to accrete multiple views of a single object.

Parallels exist between the functioning of our vision and photography or linear perspective, but because our vision exists not only in the light that enters our eyes, but also the passage of time, and the interweaving of binocular pictures of the world by our brains in conjunction with our mental image of what we expect to see, the parallel becomes problematic. Artistic practice that developed contemporaneously with photography, such as impressionism and cubism, in many ways reflects this difference.

Quote
When we look at the image on a flat plane from the proportionally scaled distance, the sense of perspective is identical for all focal lengths. Proportionally scaled is when we multiply the sensor size by a factor M, then we need to view it from a distance of focal length times M as well (which may get unpractical for ultra wide-angle and telelens shots).

The whole concept is formally called anamorphic projection distortion, or anamorphosis.

This all sounds like excuses and rationalisation to me. It's like you are saying that lens focal length shouldn't have any effect on our sense of perspective from a given position, but unfortunately it does for the following reasons.

I would almost be prepared to accept a statement that in an ideal world using perfect lenses with no distortion, whether anamorphic or otherwise, focal length would have no effect on our sense of perspective. But even that seems to be plain wrong to me.

Ignoring the anamorphic problems of ultra-wide angle lenses, subjects taken with a telephoto lens look closer. How could anyone deny it? The same scenes shot with a wider-angle lens look different and are different. They contain more subject matter. Even if one uses DXO to reduce anamophisism, they are still significantly different images in content and appearance to a telephoto shot, and as a result impart a different sense of perspective to the viewer.

I'm just reporting on what I see.  ;D
Title: Re: Does a photo give spatial information (the nose job)?
Post by: AlfSollund on January 24, 2012, 01:44:02 pm
Are you really contending that, in a photo, if one object partially obscures another, one cannot conclude that the partially obscured object is further from the camera?  Imagine for example, a photograph of a person standing in front of, and partially obscuring the lower part of a tree.  Viewing only the photo I suppose one could reach alternate conclusions:  The person is in front of the tree (closer to the camera) OR the tree is growing out of the persons head.  So granted, a photo cannot give any information about spatial relations in the real world if the observer has no experience with the real world.  In what galaxy do you reside?
Yes, I am. Viewing the photo you are correct in you conclusions. But there are other conclusions. I have already stated one; this is a photo of an photo. The object is only one, and equally close. Another; the object could have been manipulated in such a way to appear closer.

My point is that its no way of knowing for sure, since one photo is only a 2-D. And the experience from the real world makes it easier to construct the "illusion" of 3-D from a 2-D photo since we have learned that small objects are far away, that a person who are seemingly in front of a tree is so in 3-D and so on.

You are, I assume, the same AlfSollund who posted several times in a contemporary thread on the subject of perspective.  In that thread you were in firm agreement with the majority on the meaning of perspective and the fact that it depended on distance from observer to subject, and not on the focal length of the lens.  You posted:

I assume you are also in agreement with the common definition of perspective; that it involves visual distortions in a 2D image which provide information about the relative spatial positioning of objects in the original 3D scene.  Here again it is critical that the observer is cognizant of the significance of such visual distortions based on what he has seen and experienced in the real 3D world.

Now, in a separate thread, you are essentially arguing that perspective does not exist.  To paraphrase, you are arguing that "a photo cannot give any spatial information.  That we cannot derive from a photo any kind of spatial information about distances and objects spatial placement."

From this abrupt turnaround I can only conclude that you are badly confused, are unable to understand the meaning of the words (yours and others) that are being used or, perish the thought, you are simply trolling. :'(

Yes I am Alf Sollund. And I agree "perspective and the fact that it depended on distance from observer to subject, and not on the focal length of the lens". And of course perspective exist. So if you go from 3-D to 2-D you get the perspective. Buts its no going the other way since one photo only contains 2-D information.

Or to put another way. I can create the same 2-D photo with a given perspective from various 3-D scenarios (some that I "fake"). But if I take one 3-D scenario I will get one 2-D perspective from a given distance. And another perspective from another distance for the same 3-D scenario. If you combine these two you will get some spatial information. Commonly used by rangefinders .

You might argue that in taking pictures in a real world situation with many objects this is difficult to "fake". This is not what I'm arguing, please read post again. And again; I'm not arguing that its possible to make educated guesses.

Im sorry that you take this for trolling, or in any way fells offended (but how I cannot know). You might assume anything you like based on my posts about my mental health or otherwise, but I dont think this will add much to discussion  :D
Title: Re: Does a photo give spatial information (the nose job)?
Post by: AlfSollund on January 24, 2012, 02:19:47 pm
I'm sorry I haven't been able to keep up with discussion and might have missed something. Thanks to all answers.  As far as I see hjulenissen has got it correct (but you still need more than one photo). Also thanks to BartvanderWolf for his "big noise" explanation, far better than I could have done.

I think this proves that we are bound by perceiving data from a 3-D + time reality. As soon as we capture this into a 2-D photo we loose some information, but our brains are remarkably adept of reconstructing and guessing  based on what we see. This of course makes it easier to create the illusion of 3-D from a photo. A knowledge all figurative artists have been and are using.

Also, you can use different 3-D situations and get the same 2-D photo. If you assume one 3-D situation you will get one 2-D perspective independent of focal length as long as you don't change position.

P.S.
I think some are using the word "perspective" about how we compare what we see the real world (3-D by our eyes/brain), and what we see in a 2-D photo? And perhaps claiming that there is a difference by these two, and this difference is depending on focal length given the same sensor size, i.e that we are comparing two different photo "perspectives" with realty? For me its not. Its one perspective from one distance, but you can take different crops of this with a tele and wide and compare. The perspective will be the same. The comparison with reality is another discussion.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: hjulenissen on January 24, 2012, 03:51:26 pm
(http://visualfunhouse.com/wp-content/uploads/2007/09/mc_escher_063.jpg)

What is the "correct" 3-d interpretation of images such as this?

The answer of course is that several different physical "realities" can generate such an image, and humans (or dsp) cannot always reliably tell which from a single image. I think that we often can, though.

If we had information about the PDAF contrast of each of 9 or 81 AF sensors, we could be a lot more confident...

-h
Title: Re: Does a photo give spatial information (the nose job)?
Post by: mouse on January 24, 2012, 05:38:21 pm

(1) The standard lens, 50mm for 35mm format, most closely matches the natural perspective of the human eye.


This statement suggests to me that your concept of perspective remains different from that of mine, and of most others.  You will indeed find several references that claim a lens with a focal length approximately equal to the diagonal dimension of the sensor (or film, thus approx. 45mm on 35mm format) produces an image that most closely replicates the visual field of the human eye.  However visual field or angle of view is not synonymous with perspective.  Angle of view is a quality inherent in a lens; perspective is not.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: mouse on January 24, 2012, 08:49:02 pm
And of course perspective exist. So if you go from 3-D to 2-D you get the perspective. Buts its no going the other way since one photo only contains 2-D information.

Sorry, I don't understand this statement.  Are you saying that perspective exists only when we are viewing the actual 3D scene, but that in a photo perspective no longer exists?

Quote
Im sorry that you take this for trolling, or in any way fells offended (but how I cannot know). You might assume anything you like based on my posts about my mental health or otherwise, but I dont think this will add much to discussion  :D

My apologies.  I had no intent to impugn your mental health.  I was merely expressing my frustration with our failure to communicate in a common language.  Which brings me again to your failure to provide a precise definition of spatial information.  Let me try to do this for you: 

If your underlying meaning is that, given only the information contained in a photograph of a 3D scene, one cannot reconstruct a precise model of that 3D scene, where the size and position of each object is an exact replicate of the original 3D scene; then I would certainly agree with you.  However this is so restrictive a definition of spatial information that I doubt you will find many who can accept it without qualification.

To my mind, the inability to construct such an exact model is a long way from simply being able to infer some spatial information about the objects in the 3D scene simply by looking at a photo.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: hjulenissen on January 25, 2012, 02:49:56 am
To my mind, the inability to construct such an exact model is a long way from simply being able to infer some spatial information about the objects in the 3D scene simply by looking at a photo.
A full 3-d model is clearly usually not possible - any occluded objects cannot be estimated from a single 2-d image.

You certainly can "guess" many things about distance from visually inspecting a 2D image. Let's call this 2.5d (luminance+distance in each pixel) And most of the time this guess may be fairly accurate. But the estimate will sometimes be very wrong - it is not bulletproof. This is easily shown by a number of "difficult cases".

Most people (non-photographers) would probably guess that the image below is a small scaled model because we are subconsciously trained that such apparent DOF only happens when you are doing macro work.
http://farm4.staticflickr.com/3008/2952613447_53906f2f6a_o.jpg
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ray on January 25, 2012, 03:17:55 am
This statement suggests to me that your concept of perspective remains different from that of mine, and of most others.  

I don't thinks so. I believe I uderstand your concept of perspective. Let me try and describe it.

Your concept is based upon the very well-understood principle, if an object moves in relation to another object, the distance between those two objects has changed. For example, if I move my pencil closer to my notebook I can state categorically that the distance between my pencil and my notebook has been reduced.

If we imagine a robot eye that is so small it is effectively a point, smaller than a pin-head, manufactured say through nanotechnology, and imagine that robot constantly emits millions of rays in all directions as it moves around like a fly, measuring precise distances to all surrounding objects, we can state categorically that no movement equates to no change in distance and no change in perspective, and that any movement in any direction affects the distances between the robot eye and all surrounding objects, and therefore represents a change in perspective.

This is what you mean by perspective, isn't it? A totally theoretical, mathematical and abstract model which has excluded the experience of the human observer when peering through different focal lengths of lenses which results in a different 'view' of the world due to the effects of magnification or size-reduction in relation to the 'normal' perspective that a standard lens provides, or which our own eyes provide.

Quote
You will indeed find several references that claim a lens with a focal length approximately equal to the diagonal dimension of the sensor (or film, thus approx. 45mm on 35mm format) produces an image that most closely replicates the visual field of the human eye.

Hey! We almost agree on something ;D . However, I'm not at all sure that the 'standard lens' actually replicates the visual field of the human eye. I think it has more to do with degree of magnification. The field of view of the human eye seems to be vastly greater than the FoV of a 50mm lens on a 35mm format camera. As I raise the camera to my eye to view the panoramic scene out my apartment window, using my 14mm Nikkor lens on my D700, I get the impression that my view with naked eye of the same scene with fixed gaze, standing in the same position and looking in the same direction, is actually wider.

Checking on the internet for the Field of View of the naked eye, I see there's some considerable variation in results, but all results are considerably wider than the 39 degrees horizontal FoV of a standard 50mm lens.

Wikipedia quotes the FoV for simultaneous visual perception, which presumably means without moving head or eyeballs, as 160 degrees x 175 degrees. I believe that's wider than 12mm.

If one wishes to make the point that the angle of focussed view is much narrower, it is indeed, much narrower than the FoV of a standard lens. I conclude therefore that the FoV of the standard lens does not match the angle of focussed view of the naked eye, nor does it match the total FoV of the eye which includes impressions of shapes and sizes, and some sense of color, but greatly lacks detail because focussing is so narrow.

What I conclude from all this is that the appearance of the size of objects as seen by the human observer greatly influences our sense of our own distance to the object viewed. If we increase that apparent size of any object, through magnification, through viewing the object directly through a telephoto lens, or viewing a print subsequently made, the object appears closer than it actually is. With a 50mm lens, it looks about right, enabling us to fairly accurately estimate the true distance from the photographer to the subject.

Quote
However visual field or angle of view is not synonymous with perspective. Angle of view is a quality inherent in a lens; perspective is not.

You're getting a little bit philosophical here, aren't you!  ;D  I can't disagree with this statement. Perspective is a quality, a sense, an experience, an impression that can exist only in the mind of a sentient observer. It's not a quality of the lens, which is an inanimate object. A lens doesn't have a perspective, but it can change our impression or sense of perspective, if it's not a standard lens.

Interestingly, even Nikon seem to agree with me. Here's what they have to say on the subject of perspective at this website:  http://imaging.nikon.com/history/basics/19/03.htm

I've highlighted key words to help avoid any confusion in the article.

Quote
Perspective

Perspective is a term for a visual effect that causes objects to appear smaller as their distance from the viewer increases. In photography, you can control perspective by changing the lens focal length. Lenses are typically described as being either wide (wide-angle lenses) or long (telephoto lenses). Wide-angle lenses have wide angles of view, increasing the difference in the apparent sizes of objects that are at different distances from the camera. In other words, nearby objects appear larger and distant objects smaller. Telephoto lenses, in contrast, have small angles of view, decreasing the difference in the apparent sizes of objects that are at different distances from the camera. This effect can be used to exaggerate or reduce the effects of perspective by changing the focal length of the lens.

You'll note the emphasis on appearance. No-one is claiming that changing focal length of lens changes the actual and real distance to any object. That would be absurd.

Title: Re: Does a photo give spatial information (the nose job)?
Post by: AlfSollund on January 25, 2012, 05:01:25 am
Sorry, I don't understand this statement.  Are you saying that perspective exists only when we are viewing the actual 3D scene, but that in a photo perspective no longer exists?

My apologies.  I had no intent to impugn your mental health.  I was merely expressing my frustration with our failure to communicate in a common language.  Which brings me again to your failure to provide a precise definition of spatial information.  Let me try to do this for you: 

If your underlying meaning is that, given only the information contained in a photograph of a 3D scene, one cannot reconstruct a precise model of that 3D scene, where the size and position of each object is an exact replicate of the original 3D scene; then I would certainly agree with you.  However this is so restrictive a definition of spatial information that I doubt you will find many who can accept it without qualification.

To my mind, the inability to construct such an exact model is a long way from simply being able to infer some spatial information about the objects in the 3D scene simply by looking at a photo.
No offence taken, please forget this small interlude  ;).

Summary: One 3-D view translates to one 2-D perspective only. If you change position you get another 2-D perspective. But its not possible to translate the 2-D perspective back to one 3-D only, because there will be infinite 3-D models possible that led to the 2-D photo. In this sense the 2-D photo gives no spatial information. If you combine two ore more photos with different perspectives you get spatial information.

Im might not be to clear. What Im trying to say is that we see the perspective from a photo, but there is no way of translating this back to a 3-D. We can make good guesses, but its not possible to know for sure. Hjulenissen has given one example. Let me try another. let assume that I take a photo of a object, lets say a sphere towards a neutral background. From this photo it will not be possible to know if this was a sphere identical in proportions to the one in the photo but larger and placed further back.  So its possible to create the same photo perspective from these two simple scenarios. You might argue that in real life this is more difficult, but the principle still remains.

Another example. I take a photo of an mirror from an angle so that I don't show in the photo. All objects are at my back. How can you know from the photo?

So from one photo one can reconstruct infinite number of precise models of 3D scenes that corresponds to the photo.  And this is what we mentally do when seeing a photo; we create one 3-D model in our brain that fulfills what we think we see, but this is all. Its still just a guess  ;D. And so how can one argue then that the photo contains spatial information? This is just what it is, if one remove one dimension this information is lost.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Bart_van_der_Wolf on January 25, 2012, 06:28:55 am
What I conclude from all this is that the appearance of the size of objects as seen by the human observer greatly influences our sense of our own distance to the object viewed. If we increase that apparent size of any object, through magnification, through viewing the object directly through a telephoto lens, or viewing a print subsequently made, the object appears closer than it actually is. With a 50mm lens, it looks about right, enabling us to fairly accurately estimate the true distance from the photographer to the subject.

Hi Ray,

And he reason it looks about right with a 50mm lens is again due to anamorphosis. When we shoot with a 50mm lens on a 24x36mm sensor, then we should look at a print of 240x360mm at 500mm distance to get the exact same undistorted perspective 'look' as human vision perceives.

It's really more simple than people tend to realise ... Focal length equates to magnification, sensor dimensions equate to field of view (within the boundaries of the image circle), the position of the lens entrance pupil determines perspective, output viewing distance determines perspective distortion (with a planar projection, i.e. flat sensor to flat output).

Cheers,
Bart
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Ray on January 25, 2012, 11:32:29 am
Hi Ray,

And he reason it looks about right with a 50mm lens is again due to anamorphosis. When we shoot with a 50mm lens on a 24x36mm sensor, then we should look at a print of 240x360mm at 500mm distance to get the exact same undistorted perspective 'look' as human vision perceives.

It's really more simple than people tend to realise ... Focal length equates to magnification, sensor dimensions equate to field of view (within the boundaries of the image circle), lens entrance pupil determines perspective, output viewing distance determines perspective distortion (with a planar projection, i.e. flat sensor to flat output).

Cheers,
Bart


Hi Bart,
I wasn't aware that a 200mm lens (35mm format) has more anamorphosis than a 50mm lens. A 14mm lens certainly does. However the DXO converter does a reasonably good job correcting that. If one doesn't have the DXO converter with the appropriate lens module, one can simply select the problem areas with the rectangular marquee tool in Photoshop and use 'free transform'. This results in an alteration of the aspect ratio of the image, so instead of 3:2  it may become 4:3.

Quote
It's really more simple than people tend to realise ... Focal length equates to magnification, sensor dimensions equate to field of view (within the boundaries of the image circle), lens entrance pupil determines perspective, output viewing distance determines perspective distortion (with a planar projection, i.e. flat sensor to flat output).

There's some evidence that resolution and texture have a bearing on the sense of perspective. For example, if someone is presented with two images of same FoV and subject, but one of the images is a crop from a wide-angle lens and therefore low resolution and unclear, the appearance of fuzziness and lack of detail will tend to be interpreted as additional distance from viewer to subject, whereas the shot taken with an actual telephoto lens which is sharp and clear will tend to give the impression that it is closer to the viewer, even though all broad dimensions and angles are the same, indicating that the theoretical and abstract perspective is the same.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: mouse on January 25, 2012, 07:35:25 pm
Summary: One 3-D view translates to one 2-D perspective only. If you change position you get another 2-D perspective. But its not possible to translate the 2-D perspective back to one 3-D only, because there will be infinite 3-D models possible that led to the 2-D photo. In this sense the 2-D photo gives no spatial information. If you combine two ore more photos with different perspectives you get spatial information.

What Im trying to say is that we see the perspective from a photo, but there is no way of translating this back to a 3-D. We can make good guesses, but its not possible to know for sure.

So from one photo one can reconstruct infinite number of precise models of 3D scenes that corresponds to the photo.  And this is what we mentally do when seeing a photo; we create one 3-D model in our brain that fulfills what we think we see, but this is all. Its still just a guess  ;D. And so how can one argue then that the photo contains spatial information? This is just what it is, if one remove one dimension this information is lost.

Alf-

Thank you for this very clear reply.
This is a very clear statement of your meaning and I agree completely that, given your definition of spatial information, your conclusions are perfectly correct.  For you it's like pregnancy; either you are or you are not; there is no such thing as "a little bit pregnant".   :) ;)

I think our difference now simply boils down to semantics.  I define spatial information less narrowly.  Any information in a photograph (perspective for example) which provides some clues that allow one to make an informed estimate (guess) about the relative position and/or size of the 3D objects, I would describe as spatial information.  Clearly it is not always guaranteed to be correct information, and certainly one can have photos which contain no spatial information (even by this looser definition). 

I have always loved fishing.  As a very young boy I learned a trick when being photographed with my catch.  Hold the fish at arms length straight out in front of my body, and close to the camera.  At first glance it makes the fish look much bigger, but it doesn't fool too many viewers.  Which brings us back to the original theme of this thread.  It is this same "trick" that makes a nose look much bigger in proportion to the face when the camera is positioned quite close.
Title: Re: Does a photo give spatial information (the nose job)?
Post by: AlfSollund on January 26, 2012, 03:37:42 am

I think our difference now simply boils down to semantics.  I define spatial information less narrowly.  Any information in a photograph (perspective for example) which provides some clues that allow one to make an informed estimate (guess) about the relative position and/or size of the 3D objects, I would describe as spatial information.  Clearly it is not always guaranteed to be correct information, and certainly one can have photos which contain no spatial information (even by this looser definition). 

Yes, agree. And thats why in my first post wrote "The information can be used to guess about spatial, but no information can be derived for sure. Its no way of knowing the relative spatial positions to objects in a photo, or even if they are spatial." So I would have said "information to create spatial models" instead of "spatial information". But again, this might be to formal. It boils down to as you say semantics.

This is in agreement with what bill t., BJL, Ellis Vener, Hjulenissen, yourself and perhaps others have said here. Thanks to all.


I have always loved fishing.  As a very young boy I learned a trick when being photographed with my catch.  Hold the fish at arms length straight out in front of my body, and close to the camera.  At first glance it makes the fish look much bigger, but it doesn't fool too many viewers.  Which brings us back to the original theme of this thread.  It is this same "trick" that makes a nose look much bigger in proportion to the face when the camera is positioned quite close.

Totally OT, but you should look my site http://flyalf.com/ (http://flyalf.com/). On the right hand side there are links to some of my fishing trips  ;).

Check out this  ;D 8) (no, not myself in the pic):
http://www.flickr.com/photos/25078442@N05/4922442239/in/set-72157624795885356 (http://www.flickr.com/photos/25078442@N05/4922442239/in/set-72157624795885356)
Title: Re: Does a photo give spatial information (the nose job)?
Post by: Wayne Fox on January 26, 2012, 05:35:19 am

So how will we do this if I take a photo of a photo? There is only one object in this photo, and that is a (near) 2-D object. namely the photo i photograph. Secondly there are methods to guess the actual size, but not possible to know for sure.
mmm, I didn't see some of this in your original post, you didn't say anything about a photo of a photo or other qualifications.  I assume spatial information has to do with where things exist in the 3 dimensional reality of a scene in relationship to each other.  If you have specific information about the objects in the photograph it seems pretty straightforward to make measurements of the items in the photograph itself and derive some spacial information.  If there were 3 different balls in the scene ... a baseball, a football and a basketball, and they were all at different distances from the camera, you could calculate where they were in relation to each other by measuring the actual size.  Using information derived from this it seems you could calculate information about other objects in the scene where you don't know the actual size. Not sure how accurately, maybe sometimes very accurate and other times more like an educated guess.

But then I'm no scientist or math whiz, and if someone actually  asked me to do something like this, even my curiosity wouldn't be tempted by this problem and I'd find some excuse to not do it.