Luminous Landscape Forum
Equipment & Techniques => Cameras, Lenses and Shooting gear => Topic started by: Edalongthepacific on July 18, 2012, 10:10:08 pm
-
It is hard to wrap my brain around the fact that my small camera has a 7.6mm x 5.7mm CDD sensor and my larger camera has a full frame 36mm x 24mm CMOS sensor that is 20 times bigger yet not nearly 20 times better at producing quality images. I believe that the only difference in digital images is the number of pixels and the ability of the camera to deliver the most suitable color for each pixel (post-production aside). Is this true?
-
It is hard to wrap my brain around the fact that my small camera has a 7.6mm x 5.7mm CDD sensor and my larger camera has a full frame 36mm x 24mm CMOS sensor that is 20 times bigger yet not nearly 20 times better at producing quality images. I believe that the only difference in digital images is the number of pixels and the ability of the camera to deliver the most suitable color for each pixel (post-production aside). Is this true?
No, it is not.
You are leaving aside key sensor characteristics like Dynamic Range (probably the most important of all), Signal to noise ratio in mid tones, blooming behavior, ability to stream its output with sufficient speed and without heating (live view), long exposure noise,...
Cheers,
Bernard
-
Tell me more, the part after the... Also what is the difference between dynamic range and tonal range? Thanks
-
If one starts, as you did, with the assertion that the only difference between digital images (and by extension, the different sensors of your thread topic) is number of pixels and color rendition, then I would suggest you start by understanding dynamic range and signal-to-noise ratio, and leave other aspects aside for now. There have been extensive discussions of these subjects on this discussion board.
-
Hi,
The article below says it all:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/
In general, a larger sensor can record more photons. That translates into less noise, or better signal to noise ratio.
Best regards
Erik
If one starts, as you did, with the assertion that the only difference between digital images (and by extension, the different sensors of your thread topic) is number of pixels and color rendition, then I would suggest you start by understanding dynamic range and signal-to-noise ratio, and leave other aspects aside for now. There have been extensive discussions of these subjects on this discussion board.
-
Well, I think my point is being missed here. A 13 megapixel image has 13 megapixels whether it is taken with a $6000 camera or a $200 camera. The color of each pixel is chosen, for better or worse, by the camera and lens. If the perfect color is chosen for each pixel a perfect picture is the result. It does not matter what the dynamic range of the sensor is if the operator selected the wrong exposure. The picture is the sum total of camera and lens, style aside.
-
Well, I think my point is being missed here. A 13 megapixel image has 13 megapixels whether it is taken with a $6000 camera or a $200 camera. The color of each pixel is chosen, for better or worse, by the camera and lens. If the perfect color is chosen for each pixel a perfect picture is the result. It does not matter what the dynamic range of the sensor is if the operator selected the wrong exposure. The picture is the sum total of camera and lens, style aside.
Your forgetting about physics, the larger the sensor area (or photo site area) the more photons that it can capture in a given amount of time, so in general the larger the sensor the lower the noise and the greater the overall DR and IQ. If this were not the case point and shoots would create prints as nice as MFDB and we know this is not true. Dynamic range is different than exposure, it does matter, until the latest sensors no matter the exposure you could not capture what the eye can see in a single shot. I would suggest as others more reading on the subject.
Marc
-
If the perfect color is chosen for each pixel a perfect picture is the result.
Alas, this valley of tears ain't no perfect world, is it? ;D
There are multiple reasons not to record that "perfect" color, some inherent to the way light is spreading the information belonging on one pixel on its neighbours giving blur or color fringes (see eg http://www.normankoren.com/Tutorials/MTF.html to delve into the matter), others due to the unpredictable amount of noise added to the information when the sensor gathers light (see the excellent article by Emil Martinec linked above in Erik's message) ; and let's let aside the problems inherent to color theory and color reconstruction by now.
-
Well, am saying that, artistic rendering aside, image quality is a function of the selection of pixel color. The only difference between the output of a $300 sixteen megapixel digital camera and a sixteen megapixel flagship DSLR is the instrument's selection of color for each pixel. Obviously a large sensor with considerable dynamic range will produce a better image than a small sensor with a limited dynamic range, but ultimately, it is the instrument's selection of a color for each pixel of the output file that defines quality.
-
Well, am saying that, artistic rendering aside, image quality is a function of the selection of pixel color. The only difference between the output of a $300 sixteen megapixel digital camera and a sixteen megapixel flagship DSLR is the instrument's selection of color for each pixel. Obviously a large sensor with considerable dynamic range will produce a better image than a small sensor with a limited dynamic range, but ultimately, it is the instrument's selection of a color for each pixel of the output file that defines quality.
Great, you have found a great way to save tons of cash that none of us DSLR users had thought about!
Cheers,
Bernard
-
some times I wonder if Michael logs in under different names, makes silly posts just to get our goat?
Marc :)
-
"image quality is a function of the selection of pixel color"
So that's it, then? Just colour?
Noise doesn't matter, dynamic range doesn't matter (so shadows can look like crap, highlights can blow into featureless blobs of 255/255/255), sharpness doesn't matter, detail doesn't matter?
I dunno - are you being deliberately provocative, Ed? Pretty much every significant improvement to camera technology in the last few years that directly impacts on image quality, has related to one or more of the parameters I mention - none have played directly to "colour".
If you really believe what you're writing, then man! You need to get out more...
-
Well, am saying that, artistic rendering aside, image quality is a function of the selection of pixel color. The only difference between the output of a $300 sixteen megapixel digital camera and a sixteen megapixel flagship DSLR is the instrument's selection of color for each pixel. Obviously a large sensor with considerable dynamic range will produce a better image than a small sensor with a limited dynamic range, but ultimately, it is the instrument's selection of a color for each pixel of the output file that defines quality.
In one sense you are correct. If two images have the same number of pixels, the only difference between them will be the quality of the pixels. The larger, more expensive camera will usually produce better pixels for a number of reasons:
Less noise. You can think of noise as a random difference between the recorded brightness of each pixel and its "correct" value. The smaller sensor will generally have more noise, so the pixels will deviate more from their correct values.
Better dynamic range. Limited dynamic range can make it impossible to accurately capture both the brightest and darkest areas of an image with a single exposure. If you make the exposure long enough to capture the detail in the shadows, the highlights will be recorded as pure white. If you make the exposure short enough to preserve the highlights, the detail in the shadows will be lost amongst the noise. A larger dynamic range reduces this problem, although all current sensors (and the human eye) will have difficulty with some scenes.
A reduction in various other artifacts (blooming, etc.) which I don't know enough about to comment on.
Larger pixels allow you to use a narrower aperture before diffraction starts to blur the image. Some small, high pixel count cameras are limited by diffraction even with the lens wide open.
On the other hand, smaller sensors do have one advantage - more depth of field for a given aperture and field of view. This can also be a disadvantage if you want a shallow depth of field for artistic reasons.
-
Very well said. Thank you all.
-
In general, a larger sensor can record more photons. That translates into less noise, or better signal to noise ratio.
I've never really understood or agreed with that view.
No matter what size the sensor is the number of photons which are recorded at each individual pixel (assuming for a moment for easy of argument that pixels are all the same size no matter what size the sensor) will be the same regardless of sensor size. All the larger sensor does is receive more photons in the area which is beyond the physical area of the smaller sensor.
Imagine a brick wall facing the sun, each brick illuminated equally. Now draw two squares on it, one within the other and each enclosing a number of bricks. The larger square will capture more light but the quality and strength of the light falling on each brick is exactly the same. Bricks / light outside of any square we draw can not alter the amount or quality of light inside of the square. You could prove this by nailing a light meter to each brick. Adding or removing bricks will not affect the light meter readings on other bricks. How could it.
Back to our sensor... The size of the pixel must matter to the signal to noise ratio and other aspects of image quality as must the physical make up of the track and its length to the A/D converter / other components. The other clever electronic and algorithms must also matter. The problem we have when comparing different sized sensors is that there are other differences than sensor size. Pixel size varies, the basic design of the chip varies with some having on board A/D and some having off board A/D, the length of track varies, the shape of tracks varies, firmware varies etc...
But wait... we do have some cameras that have a crop mode... If we take a FF shot and then engage the crop mode will DR and noise automatically get worse? I don't know for sure because I've never done it but I suspect that all would remain the same as all we are doing is throwing away or not capturing photons beyond our image just like we do when we crop an existing image.
Back to the original question. The size of pixels must matter but it's more than that. The design of the chip and circuit beyond pixel size must matter as must the programming. The whole chain must contribute to the final image quality.
-
If this were not the case point and shoots would create prints as nice as MFDB and we know this is not true.
But the compact camera sensor has teeny tiny pixels and was made for 3p from a design someone did on a fag packet whilst the MFDB has massive pixels and was lovingly designed by a huge team of hand picked guys with a big budget. No wonder it's better.
-
Actually if you read the technical discussions on this site both photosite size and total sensor area affect the IQ of the total Image
Remember DxO is only rating the pixel quality not the quality of the sensor as a whole. It's not just resolution but as the size of the sensor increases the total noise decreases and the total DR increases.
Too often we evaluate a 100% crop and forget about the entire image. Per pixel my D800E is the highest IQ camera that I have, but per picture my IQ180/Rodenstock HR's is superior.
Marc
-
Actually if you read the technical discussions on this site both photosite size and total sensor area affect the IQ of the total Image
How can this be true though? How can total chip area affect the image and if it is true we'll see a drop in IQ when we crop an image by switching off / masking off an area of the sensor. I see no drop in IQ when I crop an image and I wouldn't expect a FF camera with a built in crop mode to give lower IQ in that mode.
Personally, until someone demomstrates to me is person that chip size in itself affects IQ I probably wont belive it :)
PS. My background is about thirty years of computing and wider electronics repair and technology products manufacturing, not cameras :) That may make my views invalid :) but I'll wait to be convinced :)
-
Personally, until someone demomstrates to me is person that chip size in itself affects IQ I probably won't believe it :)
Does a big window let in more light than a small window?
That's a perfect - if somewhat simplistic - analogy for the big sensor/small sensor question: other things being equal, more light hitting the sensor equals better SNR equals better IQ.
-
No matter what size the sensor is the number of photons which are recorded at each individual pixel (assuming for a moment for easy of argument that pixels are all the same size no matter what size the sensor) will be the same regardless of sensor size. All the larger sensor does is receive more photons in the area which is beyond the physical area of the smaller sensor.
This thread was about comparing sensors with the same number of pixels and different sizes.
Sure, if sensor B has 1/4 the area and 1/4 the pixel count of sensor A, and both are built with equivalent technology, the per-pixel quality should be about the same. However, sensor A will have double the resolution, assuming that the lens is adequate. Photos from the larger sensor will show less noise and other artifacts for a given print size.
Back to the original question. The size of pixels must matter but it's more than that. The design of the chip and circuit beyond pixel size must matter as must the programming. The whole chain must contribute to the final image quality.
That's pretty much stating the obvious, and I don't think anyone could disagree with it. Size matters a lot, but obviously it's not the only thing that matters. For example, I would expect the best micro four-thirds sensors to produce better images than an old full-frame sensor like the one in the Canon 1Ds. It would be an interesting comparison if anyone has access to the relevant cameras.
-
How can this be true though? How can total chip area affect the image and if it is true we'll see a drop in IQ when we crop an image by switching off / masking off an area of the sensor. I see no drop in IQ when I crop an image and I wouldn't expect a FF camera with a built in crop mode to give lower IQ in that mode.
Of course you see no drop in pixel IQ when you crop an image - they're the same pixels. But it's not the same picture. To get the same picture (same composition and angular coverage) from that cropped subset of pixels, you'd have to mount a shorter focal-length lens. Now you have a clear and obvious drop in image IQ - lower magnification, lower angular resolution, lower photon count per object within the photo (something which covered say 9 pixels in the original now only covers 4 pixels, if you employed a 1.5x crop factor).
Personally, until someone demomstrates to me is person that chip size in itself affects IQ I probably wont belive it :)
I hope you believe it now? I think you just had to disentangle pixel IQ from overall image/picture IQ in your mind.
Ray
-
It is hard to wrap my brain around the fact that my small camera has a 7.6mm x 5.7mm CDD sensor and my larger camera has a full frame 36mm x 24mm CMOS sensor that is 20 times bigger yet not nearly 20 times better at producing quality images. I believe that the only difference in digital images is the number of pixels and the ability of the camera to deliver the most suitable color for each pixel (post-production aside). Is this true?
Ed(?) - your belief is pretty far off the truth, but you are at least asking the right questions and looking for the truth so I won't knock that.
Your question is a perfect candidate for a reductio ad absurdum way of answering. (This is a logical technique that you can employ yourself of course). If size doesn't matter, then let's take that 7.6mm x 5.7mm sensor and shrink it by a factor of 1000, so now it's a 7.6 micron x 5.7 micron sensor. And let's take that 36mm x 24mm sensor and enlarge it by a factor of 1000, so now it's a 36 metre x 24 metre sensor. Both still have the same number of pixels as before. Do you still think that they will give comparable or even same-ballpark image quality?
Well, am saying that, artistic rendering aside, image quality is a function of the selection of pixel color. The only difference between the output of a $300 sixteen megapixel digital camera and a sixteen megapixel flagship DSLR is the instrument's selection of color for each pixel.
That's twice you use the expression "selection of color" - "selection" is an unfortunate term because it implies that the camera has some choice, some informed decision to make about each pixel. It doesn't. It doesn't select or choose, it simply measures. (It also doesn't work with or measure "color" - it measures greyscale intensity at each pixel - the color comes afterwards, by interpolation across adjacent pixels which captured the light through different built-in coloured filters).
What distinguishes your two different sixteen megapixel sensors is how accurately they can measure the underlying rate of light flux falling on each pixel. You can check the references given above for a description of the Poisson nature of photon arrival times. To keep it simple, all I'll say is that the sensors only have a limited sample of that light flux to work with. The more light each pixel captured, the more likely that its measurement is closer to the true underlying rate of light flux. Given equal sensor technologies, this means that a larger 16MP sensor gives better IQ than a smaller 16MP sensor, due to its larger pixels capturing more light in a given exposure time. This scales nicely even when you apply the reductio ad absurdum test. At the microscopic end of the absurdum, you eventually hit the limits of discrete Poisson stats and quantum electronics.
Keep asking yourself such questions, keep researching and learning. In this age of depressingly rampant black-boxism ("it just works, I have no idea why"), you'll still find lots of people willing to help.
Ray
-
Quote from Cambridge in colour
"even if two sensors have the same apparent noise when viewed at 100%, the sensor with the higher pixel count will produce a cleaner looking final print. This is because the noise gets enlarged less for the higher pixel count sensor (for a given print size), therefore this noise has a higher frequency and thus appears finer grained."
"No matter what the pixel size, larger sensors unavoidably have more light-gathering area. Theoretically, a larger sensor with smaller pixels will still have lower apparent noise (for a given print size) than a smaller sensor with larger pixels (and a resulting much lower total pixel count). This is because noise in the higher resolution camera gets enlarged less, even if it may look noisier at 100% on your computer screen. Alternatively, one could conceivably average adjacent pixels in the higher pixel count sensor (thereby reducing random noise) while still achieving the resolution of the lower pixel count sensor. This is why images downsized for the web and small prints look so noise-free."
http://www.cambridgeincolour.com/tutorials/digital-camera-sensor-size.htm
Marc
-
Of course you see no drop in pixel IQ when you crop an image - they're the same pixels. But it's not the same picture. To get the same picture (same composition and angular coverage) from that cropped subset of pixels, you'd have to mount a shorter focal-length lens. Now you have a clear and obvious drop in image IQ - lower magnification, lower angular resolution, lower photon count per object within the photo (something which covered say 9 pixels in the original now only covers 4 pixels, if you employed a 1.5x crop factor).
I hope you believe it now? I think you just had to disentangle pixel IQ from overall image/picture IQ in your mind.
Ray
No. Because your argument is nothing to do with sensor seize affecting IQ in itself and by itself. Yes, a bigger chip will enable you to print a bigger picture but that isn't image quality...
-
Quote from Cambridge in colour....
Imagine a sensor... Take a picture and view it. Now imagine that we can electrically isolate rows of pixels making the resultant image smaller. Some cameras can actually do this.
By the reasoning of many the IQ of the remaining image will drop. I still remain to be convinced of this and in fact I don't believe it.
Yes the smaller image will degrade if magnified more to match the size of the larger image but taking a picture with a smaller sensor should be the same as cropping an image taken with a larger system using the same technology. The IQ in the areas of the image that both systems capture should be the same at the same image size as all you are doing is throwing away the edges of the larger picture or not capturing them with the smaller system.
-
Scooby70 you are changing the argument, please reread the original post! Of course each pixel is equivalent that is why many go for apsc sized cameras when the pixels are the same, the apsc sized cameras take advantage of the sweet spot of the lens. But the original statement was comparing cameras, for instance if I had a 16mp db and took a picture and say compared the same picture to a 16mp point and shoot camera is there an IQ difference? the OP thought the only difference was how they interpret color. This is not true the larger sensels and the larger sensor will in general have higher IQ.
Marc
-
Hi,
I'm looking into this a bit, right now. I happen to have three cameras with similar resolution but different sensor sizes.
Sony Alpha 900, full frame, 24MP
Sony Alpha 77SLT, APS-C 24 MP
Sony RX100, small sensor, 20MP
I plan to do some kind of write up on the issue. Anyway, what I see is that the difference between the two large sensor cameras may be small. The Alpha 77 SLT has live view while the Alpha 900 has not, and that may be the most important factor between the two.
The small sensor camera can hold it's own at large apertures and short focal lengths. It is quite impressive, but it cannot really match the larger sensors. The RX100 is really good at handheld low light shooting at short focal lengths as the lens is incredibly good at full aperture!
Best regards
Erik
Scooby70 you are changing the argument, please reread the original post! Of course each pixel is equivalent that is why many go for apsc sized cameras when the pixels are the same, the apsc sized cameras take advantage of the sweet spot of the lens. But the original statement was comparing cameras, for instance if I had a 16mp db and took a picture and say compared the same picture to a 16mp point and shoot camera is there an IQ difference? the OP thought the only difference was how they interpret color. This is not true the larger sensels and the larger sensor will in general have higher IQ.
Marc
-
Charles Johnson, Science for the Curious Photographer. It does have some math, but it is written in such a way that you don't need math skills to understand most of it.
-
Yes, that is a very good book. Thank you!
Erik
Charles Johnson, Science for the Curious Photographer. It does have some math, but it is written in such a way that you don't need math skills to understand most of it.
-
No matter what size the sensor is the number of photons which are recorded at each individual pixel (assuming for a moment for easy of argument that pixels are all the same size no matter what size the sensor) will be the same regardless of sensor size. All the larger sensor does is receive more photons in the area which is beyond the physical area of the smaller sensor.
Imagine a brick wall facing the sun, each brick illuminated equally. Now draw two squares on it, one within the other and each enclosing a number of bricks. The larger square will capture more light but the quality and strength of the light falling on each brick is exactly the same. Bricks / light outside of any square we draw can not alter the amount or quality of light inside of the square. You could prove this by nailing a light meter to each brick. Adding or removing bricks will not affect the light meter readings on other bricks. How could it.
I presume your desired end product is a photograph of the inner section of the brick wall? If you take images as you have described above, both cameras will produce identical results, as you note. But your setup deliberately throws away pixels from the larger format camera. Let's change the rules so that both cameras photograph the same subject, namely the inner section of the brick wall, without cropping. This could be accomplished for example by switching to a longer focal length lens on the larger format camera. Now the larger camera gets to use its entire sensor, and the resulting image will almost certainly be superior. How could it not be better to use more pixels?
In this revised setup, if both lenses are set to the same aperture (say f/8 in both cases), then the light intensity per unit area on the sensor will be the same, so each pixel will still collect the same number of photons. The longer lens collects more light through a larger physical opening, but that light is distributed to a correspondingly larger area on the sensor. What is different in this setup is the resolution -- a single brick is imaged across more pixels on the larger sensor. This is what potentially gives the larger sensor its advantage. This resolution advantage also provides a signal/noise advantage to the overall image, assuming that noise per pixel is the same on both sensors.