Pages: [1]   Go Down

Author Topic: Fuji S3 Pro sensor analysis  (Read 8031 times)

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
Fuji S3 Pro sensor analysis
« on: January 10, 2005, 10:14:23 am »

I think that your advice about underexposure is extremely ill-advised:

Quote
These things mean that you can "underexpose" images by one or two stops (three if the brightness range of the subject is small) in order to guarantee against highlights blocking up. The price of doing so is more work in post-processing; all your images may print too dark unless manipulated. The advantage is that it's much easier to create a terrific print if you leave yourself room on both ends of the characteristic curve.

As long as clipping is avoided, it is always easier to make a good print from a file with more exposure rather than less, especially when using ISO settings where noise must be dealt with. Sometimes this requires underexposing the midtones, but if the dynamic range of the scene is less than the camera's capture ability, exposure should be set so that the gap on the right of the histogram is smaller than the left for best results.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Fuji S3 Pro sensor analysis
« Reply #1 on: January 12, 2005, 12:48:35 pm »

Quote
"Ideal exposure with a digital camera (capturing the greatest possible dynamic range with the lowest possible level of noise and and other undesirable digital artifacts at a given ISO setting) is always that which causes the RAW data to closely approach, but not reach the clipping point for non-specular highlights."
That seems sound advice in most cases, but assumes that the exposure index ("ISO") has already been chosen; say the best (usually lowest) for the sensor.

However, if a sensor has dramatically increased highlight handling capability, the lowest usable exposure index can get very low, too low to use in many situations, so one has to juggle choices of three exposure parameters: aperture, shutter speed and exposure index. You get to choose any two, then the other is determined.

In situations where the lowest usable exposure index (highest tolerable amount of exposure before significant highlights are blown out) is too low, one might instead work in terms of choosing aperture and shutter speed according to desires for DOF and freezing (or deliberately blurring) motion, and then use the exposure index dictated by that combination.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #2 on: January 14, 2005, 07:40:59 pm »

Quote
  I am referring to cameras that have substantially expanded highlight headroom, so that the "base ISO speed" drops well below the conventional 100 or so.
BJL,
Which cameras are these. I recall the base ISO of the Kodak 14n is ISO 80. I believe Canon might have a few P&S cameras with a base ISO of 50. Usually (prehaps always) the cameras with a lower base ISO than 100 also have a lower maximum ISO setting. I don't see how this changes anything.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #3 on: January 17, 2005, 07:20:08 am »

Quote
- Set the camera to its lowest ISO, lowest contrast, WB appropriate to the light source, and make small jpeg files.
Thanks for the advice, but I never use jpegs unless I'm running out of memory. As far as I understand, if people are at all concerned about getting the most DR from their camera then shooting in RAW is mandatory.

I'm not sure I need to know precisely what the DR is in terms of F stops. I was just curious as to whether or not the 20D had greater DR than the D60 in any meaningful way. It doesn't appear to have.

ps. I can see the dual pixel S3 is in a different category. If I understand the process, there has to be some sort of in-camera blending of the two readings. I take it there is no RAW option in wide mode. Is that right?
Logged

johnsondanl

  • Newbie
  • *
  • Offline Offline
  • Posts: 5
Fuji S3 Pro sensor analysis
« Reply #4 on: January 09, 2005, 09:25:42 pm »

For your viewing enjoyment, my weekend's testing with an S3, to learn whether its sensor truly offers a discernably increased dynamic range (it does, but it won't hit you in the face):

http://www.danlj.org/~danlj/photos/techniq...jiS3/index.html
Logged

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Fuji S3 Pro sensor analysis
« Reply #5 on: January 10, 2005, 02:31:09 am »

Thanks for posting. It does seem there is a problem with the standard DR image. At the top is horizontal banding visible.

I believe the dynamic range increase is actually obvious and significant, I also still believe that the particular noise inherent to Fuji's superCCD is killing all the fun. It would also be interesting to see how the DR increase holds up in different colors...
Logged
Regards,
~ O ~
If you can stomach it: pictures

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #6 on: January 10, 2005, 10:28:18 pm »

Quote
I think that your advice about underexposure is extremely ill-advised:

As long as clipping is avoided, it is always easier to make a good print from a file with more exposure rather than less
That's generally true. But there are sometimes borderline situations regarding specral highlights. The histogram is not always so precise that it can differentiate between what is a desired specral highlight and what is a bit more than a specral highlight.

In such circumstances there's an almost invisible flat line between the main body of the histogram and the right side of the histogram. Move the rightmost slider to the edge of the main body of the histogram and you've enlarged the specral highlights to beyond acceptability.

There's a fine line here. Depending on the nature of the scene, it can be good advice to pull back from an 'expose to the extreme right' scenario. Ever tried photographing white Pelicans in the sunlight?
Logged

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
Fuji S3 Pro sensor analysis
« Reply #7 on: January 11, 2005, 01:27:28 am »

Quote
Quote
I think that your advice about underexposure is extremely ill-advised:

As long as clipping is avoided, it is always easier to make a good print from a file with more exposure rather than less
That's generally true. But there are sometimes borderline situations regarding specral highlights. The histogram is not always so precise that it can differentiate between what is a desired specral highlight and what is a bit more than a specral highlight.

In such circumstances there's an almost invisible flat line between the main body of the histogram and the right side of the histogram. Move the rightmost slider to the edge of the main body of the histogram and you've enlarged the specral highlights to beyond acceptability.
High DR situations are always a PITA to expose properly; finding nirvana gererally revolves around figuring out the maximum amount of highlight clipping you can get away with and still keep decent shadow detail. That's why I generally state the following:

"Ideal exposure with a digital camera (capturing the greatest possible dynamic range with the lowest possible level of noise and and other undesirable digital artifacts at a given ISO setting) is always that which causes the RAW data to closely approach, but not reach the clipping point for non-specular highlights." (From Digital Exposure And Metering Strategies)

I neglected to add the italicized bit in my previous post. Following this advice can result in brightness adjustments of 2 stops or more sometimes being necessary in post, but this is generally not desirable if it can be avoided.

Quote
Ever tried photographing white Pelicans in the sunlight?

No, but I have shot egrets. Same deal.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #8 on: January 12, 2005, 10:22:53 pm »

Geez! BJL, you have a way of making things appear more complicated than they really are   . How about:-

(1) Always expose as far as possible to the right without blowing highlights, whatever the ISO setting.

(2) Select desired aperture and shutter for purposes of DoF and/or freeze or blur requirements.

(3) Select appropriate ISO for correct exposure.

(4) Fine tune one or more of the above 3 settings to comply more accurately with condition (1) before pressing the shutter.  
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Fuji S3 Pro sensor analysis
« Reply #9 on: January 13, 2005, 12:27:17 pm »

Quote
Geez! BJL, you have a way of making things appear more complicated than they really are   . How about:-

(1) Always expose as far as possible to the right without blowing highlights, whatever the ISO setting.
 ...
Why not? Because you have to think about the exposure index (ISO) up front with technology that greatly increases highlight headroom, corresponding to a very low EI settings. Just using lowest available EI most of time and kicking it up only when high speeds are needed will not work so well if minimum EI setting is 25, or 6 (Kodak SLR/n?) or less.

But another way to think of it is this: aperture and shutter speed have a fairly direct relationship to ones goals for the image, by affecting depth of field, motion blur and such; exposure index has no such direct aesthetic significance.

Therefore, I would find it artistically simpler to be able to choose an aperture and a shutter speed, and have exposure index chosen according to guidance from the light meter, checking the value in case it is too high for good shadow handling. Sort of "aperture AND shutter speed priority", with the EI being the quantity then set automatically.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Fuji S3 Pro sensor analysis
« Reply #10 on: January 14, 2005, 11:08:06 am »

Ray,

   I am referring to cameras that have substantially expanded highlight headroom, so that the "base ISO speed" drops well below the conventional 100 or so.  Every doubling of highlight headroom at a given photosite size halves the base ISO, and it seems quite likely that the Fuji SR idea has the potential to add more than two stops of headroom; then you could have a base ISO of less than 25, which would not be my optimal setting in most situations.

Your experiment with the 20D is interesting and I should try a similar one with the E-1; my guess is that at least at EI 200 and possibly at 400, the noise floor is low enough that it is below the print black point for straight prints. By "straight prints" I mean ones without contrast contraction or dodging of shadow details, and by "print black point" I mean the lowest attainable reflectivity, which I understand is between 1% and 2%. That is about four stops below mid-tones placed at the conventional 18%, so more than four stops of shadow depth (going below Zone I) is completely wasted on straight prints. Maybe this is a reason why the ISO film speed is defined in terms of performance four stops below mid-tones.

For me the optimal exposure index for most shots, particularly hand-held, might be the highest one with no noticable noise down to that "print black point", three of four stops below the mid-tones. Lower speeds would only be preferable when deeply shaded details are in play.

P. S. About "use with care" low EI settings like ISO 50 on the EOS-1 Ds.
I believe that the exposure index settings used on DSLRs are rather cautious about highlights, allowing for them to go about 3 1/4 stops above metered mid-tones, whereas if the mid-tones are at the conventional 18%, 2 1/2 stops is enough for everything except specular highlights. I imagine that those special low EI settings are simply giving up this margin of safety, so that they are fine in scenes with a normal to low brightness range from mid-tones to highlights.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Fuji S3 Pro sensor analysis
« Reply #11 on: January 16, 2005, 12:47:57 pm »

Quote
Quote
  I am referring to cameras that have substantially expanded highlight headroom, so that the "base ISO speed" drops well below the conventional 100 or so.
BJL,
Which cameras are these.
Perhaps more future than current. The new Kodak SLR/* models have an ultra-low minimum EI setting of 6, and I have read that the base ISO of the sensor in the earlier Kodak 14/N is lower than its minimum ISO setting, maybe as low as 40.  The Fuji S3 might well have about two extra stops of headroom, but not the correspondingly low ISO settings, so that accessing its true base ISO speed level might require methods like deliberate "overexposure".

The future prospects include the Stanford University/Pixim approach discussed in another thread recently, currently implemented only in Pixim surveillance video cameras.
Logged

johnsondanl

  • Newbie
  • *
  • Offline Offline
  • Posts: 5
Fuji S3 Pro sensor analysis
« Reply #12 on: January 16, 2005, 08:09:58 pm »

If you really and truly want to know whether one camera has greater usable dynamic range than another, do this:
- Set up a target and meter it.  Ideally, choose an f-stop around 1/4 - 1/10 sec to give yourself lots of room on both sides of the middle -- at least five stops above (longer shutter speeds) and at least 8 stops below (shorter shutter speeds).
- Set the camera to its lowest ISO, lowest contrast, WB appropriate to the light source, and make small jpeg files.
- Shoot a series of exposures from the camera's highest shutter speed to about 8 stops beyond the metered exposure (e.g., 4 - 6 seconds)
- Copy all the pix onto your computer and load your image-processing program (the gimp, Photoshop, etc.)
- Load all the images you can into this program, then use the curves tool (in color) or the eyedropper tool (in monochrome) to measure the pixel intensity of each image.
- Note visually which dark image first has "unusable" noise.
- Note visually which light image last has texture.
- Use your favorite spreadsheet program to record these values versus the relative exposure,
- Repeat this exercise with the other camera, and record its values in your spreadsheet in a second column
- then create a line chart of these values.
- Count the stops between the darkest usable image and the brightest with texture, and between that and pixel saturation.
This is the usable dynamic range.

It's actually pretty fast; about 10 minutes for all the exposures and a couple of hours for the post-processing.

For an example chart:
http://www.danlj.org/~danlj/photos/techniq...amicRange2b.jpg

Best wishes!
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #13 on: January 17, 2005, 08:14:09 pm »

Re-reading Michael's review of the S3 I see there is a RAW output option in wide mode, but it's slow and cumbersome with huge, uncompressed files and doesn't appear to offer any worthwhile increase in DR over the jpeg output. This is not the case with DSLRs like the Canon 20D which can produce up to 2 stops increase in DR by shooting RAW and using a good converter.

The quality of the converter is crucial to achieving this increase in DR from a RAW image. When I started using ACR which is packaged with PSCS, I soon realised it did a much better job at recovering highlight detail than Canon's own Zoombrowser or even BreezeBrowser, but I don't know how the latest version of BreezeBrowser stacks up.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #14 on: January 09, 2005, 10:26:59 pm »

Thanks for a very thorough analysis. The increase of dynamic range of about 2 stops in wide mode, compared to standard mode and the Konica A2, is clearly significant.

But the S3 is more expensive than the Canon 20D. This is the competition I would have thought. On another thread it's mention that the 20D has slightly more resolution. If it also has about equal DR, or as close as matters, the S3 is not looking so good.
Logged

johnsondanl

  • Newbie
  • *
  • Offline Offline
  • Posts: 5
Fuji S3 Pro sensor analysis
« Reply #15 on: January 11, 2005, 12:35:13 am »

You both make appropriate points; "extremely ill-advised" simply reflects a misunderstanding of my point.  I put "underexpose" in quotes to indicate irony: namely, that sometimes, to protect highlight texture, you must disbelieve the camera's exposure-calculation algorithm; and the two sensors I've tested have at *least* two stops of shadow latitude at their lowest ISO ratings (the S3 has three, for sure).
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #16 on: January 13, 2005, 08:45:11 pm »

Quote
Why not? Because you have to think about the exposure index (ISO) up front with technology that greatly increases highlight headroom, corresponding to a very low EI settings. Just using lowest available EI most of time and kicking it up only when high speeds are needed will not work so well if minimum EI setting is 25, or 6 (Kodak SLR/n?) or less.
BJL,
You are still not clear. Are you referring to cameras that have sub-optimal ISO settings both above and below the optimal setting which is usually ISO 100? I understand that ISO 50 on the 1Ds will compress dynamic range a little, I presume at the expense of highlight headroom, but I see no reason why the 'expose to the right' rule should change with higher ISO settings.

Just to confirm this, I stepped outside onto the sun lit verandah and took 3 shots with my 20D at f8 with the following ISO/shutter speed combinations:-

(1) 1/100th at ISO 100

(2) 1/400th at ISO 400

(3) 1/1600th at ISO 1600

The scene was exposed for the parts that were not in direct sunlight. Part of the concrete floor was in direct sunlight and this part, on conversion from RAW, was almost totally blown over the entire area in all 3 images. The only detail consisted of just a few specks. However, if one wished to put a fine point on it, there were one or 2 more visible specks in the ISO 100 image, than in the 400 image, than in the 1600 image.

I reconverted all 3 images using the full minus 2 stop exposure compensation that ACR allows, and almost all the lost detail in the concrete floor was recovered in all images except[/i] the ISO 100 image still showed slightly more detail, but more significantly some yellow stains on the concrete floor were almost lost in the higher ISO shots. In the ISO 100 shots, those yellow stains were as clear and prominent as in the real scene.

Hhmm! Fine camera this 20D  :D .
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #17 on: January 16, 2005, 05:59:09 pm »

I just checked to see if the 20D has more or less the same DR as the D60, overexposing each to the same degree, then examining shadow and highlight detail after the same amount of EC in ACR. They both appear to be very close indeed. I tend to give the edge to the 20D. For example an overexposed 1/50th sec with the 20D looks the same as an overexposed 1/60th sec with the D60, after EC on conversion. But this could be due to slight differences in the true base ISO of the two cameras.

The 20D produces a more accurate exposure due to its greater number of shutter speed options. For the same scene the D60 will use 1/180th whereas the 20D will use 1/200th. The D60 doesn't have a 200th. Getting blown highlights with the D60 using evaluative metering happens more frequently.
Logged

johnsondanl

  • Newbie
  • *
  • Offline Offline
  • Posts: 5
Fuji S3 Pro sensor analysis
« Reply #18 on: January 17, 2005, 08:38:05 pm »

Quote
Quote
- Set the camera to its lowest ISO, lowest contrast, WB appropriate to the light source, and make small jpeg files.
Thanks for the advice, but I never use jpegs unless I'm running out of memory. As far as I understand, if people are at all concerned about getting the most DR from their camera then shooting in RAW is mandatory.
Sorry.  I didn't make the point clear.  Small jpeg files to make the task quicker.  Once you go to jpeg, the DR isn't altered; in fact, you can't test the camera's image-processing software unless you make tiff or jpegs.  Shooting raw increases the saved information, but confuses the test: you are measuring the camera's sensor and whatever post-processing software being used to produce the image.  I just want to test the camera, and don't have the time to explore the great complexity of RAW and which processing choices to make in the conversion.  And I don't have any software that is able to look at the pixel values directly of the RAW image.  Nor do I know of any.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Fuji S3 Pro sensor analysis
« Reply #19 on: January 17, 2005, 10:29:21 pm »

What I did was to take a number of shots of the same scene with both cameras in RAW mode, varying each shot from clear underexposure to extreme overexposure, in increments according to the shutter speed options. The 20D has more shutter speed options so things were not quite sychronised, but I was able to match a 20D 50th sec exposure with a D60 60th sec with regard to highlight detail after equal exposure compensation during conversion, (minus 2 stops).

I then examined the deepest shadows in both matched images, lightening them further with the 'levels' slider to see if there was more shadow detail in one than the other. There wasn't, so I concluded that for all practical purposes the DR of both cameras is the same.

More careful testing might reveal more subtle differences, but I doubt that such small differences would be meaningful or relevant in the general scheme of making/taking photos.

I don't see the need to make a distinction between the camera's innate DR and its DR in conjunction with a good RAW converter. No computer hardware works without software.

I used the same lens at f8 on both cameras. Lighting condition would have changed slightly during the 5 minutes of testing. It was a sunny day with few clouds.
Logged
Pages: [1]   Go Up