Lenshacker Posted March 10, 2015 Share #261 Posted March 10, 2015 Advertisement (gone after registration) I am not sure the difference would be detectable after gamma compression. This has been Leica's argument in defense of DNG-8 for years. To verify your hypothesis, it should be enough convert a linear DNG (no postprocessing) to DNG-8. Then compress both files to JPEG and compare their size. The results: Uncompressed DNG stored as Full-sized JPEG: 7177K; reduced to 1800x1198: 1131K. Compressed DNG stored as Full-sized JPEG: 7326K; reduced to 1800x1198: 1142K. Of course the uncompressed DNG had to be converted to 8-bit to store as JPEG, but nothing else done. My "speculation": The DNG8 algorithm cuts out noise, but introduces sharp edges as the intensity step-size is coarse. The sharp edge requires higher frequency to reconstitute. Lower resolution, the edge is smoothed over- but not completely. I'll be bummed out if David used DNG-8 for his test! The compressed DNG file is made using my FORTRAN code to implement to DNG-8 functions. I verified that the M9 uses the identical table as the M8, it is stored in file offsets '0614'x through '0812'x. I wrote most of this software ~20 years ago for looking at TIFF 6.0 files that my wife needed for work. The researcher had stored multiple images in the same TIFF file, but Photoshop 3.0 could not see past image 1. Link to post Share on other sites More sharing options...
Advertisement Posted March 10, 2015 Posted March 10, 2015 Hi Lenshacker, Take a look here CCD vs CMOS: Can you tell which is which?{merged}. I'm sure you'll find what you were looking for!
Lenshacker Posted March 10, 2015 Share #262 Posted March 10, 2015 After reading a lot of comments debating the merits of CCD vs CMOS (M9 vs M240), I decided to do a comparison of the two cameras. I shot with both at the same time, using the same settings and same lenses. I then used Lightroom to do a rough match of the images using only global slider adjustments. No Photoshop. No adjustment brushes or any other local adjustments. David David- Did you use Compressed or Uncompressed DNG for the M9 images? Link to post Share on other sites More sharing options...
dfarkas Posted March 10, 2015 Author Share #263 Posted March 10, 2015 David- Did you use Compressed or Uncompressed DNG for the M9 images? Compressed DNG on both cameras. I am well aware that the DNG compression on the M9 is technically lossy. In practice, and especially in "easy" lighting scenarios like in my test, there is no discernible visual difference in quality. Am I glad that Leica switched to lossless compression? Of course. Link to post Share on other sites More sharing options...
TonyField Posted March 10, 2015 Share #264 Posted March 10, 2015 I probably cannot tell the difference between M9 compressed and uncompressed files .. I have yet to see any difference with the subject material i shoot.. The M9 manual stipulates that the M9 has a slighly lossy compression scheme. Has there been any firmware updates that give lossless compression ?? Link to post Share on other sites More sharing options...
Nordvik Posted March 10, 2015 Share #265 Posted March 10, 2015 Thanks for this test. Can you please upload some DNG-pairs? One high contrast and one low contrast? Link to post Share on other sites More sharing options...
CheshireCat Posted March 10, 2015 Share #266 Posted March 10, 2015 I am having a lot of fun with this. I was able to reconstruct most of the image from the residual error alone. I will alter the subroutine to write the truncated data to disk, rather than the residual error. Interesting. Please post pictures. Link to post Share on other sites More sharing options...
Lenshacker Posted March 10, 2015 Share #267 Posted March 10, 2015 Advertisement (gone after registration) Compressed DNG on both cameras. I am well aware that the DNG compression on the M9 is technically lossy. In practice, and especially in "easy" lighting scenarios like in my test, there is no discernible visual difference in quality. Am I glad that Leica switched to lossless compression? Of course. That explains the difference in frequency content. Just not what I thought it was. Link to post Share on other sites More sharing options...
Lenshacker Posted March 10, 2015 Share #268 Posted March 10, 2015 Interesting. Please post pictures. This is the original- I chose this one for the Highlights. L1000439 And this is generated using LightRoom Sliders from the Data that is discarded by the DNG-8 compression. I replaced all of the pixels in the DNG file with original value- the "restored Compressed" value. It is the difference frame between an uncompressed image and the image that is generated using (( Sqrt( i* 4))**2)/ 4), integer math. DNG8DIF2 I feel just like Peter Max. This is one of the first pictures that I took with the M9. I'm doing some examples for Arvid's M8RAW2DNG program. I like reading the older threads discussing the M8. Seems a number of forum members wrote a lot of code to explore the DNG files. I don't recall a comparison like this one. The first digital sensor that I worked with had a habit of getting "Stuck Bits". I wrote code that looked for the stuck bit and tried to flip it to restore the pixel. I am feeling very old right now, but it is fun to dig out code that goes back 25+ years and use it on a modern camera. Link to post Share on other sites More sharing options...
CheshireCat Posted March 11, 2015 Share #269 Posted March 11, 2015 Curioser and curioser Apart from areas with clipping, I would have intuitively expected a more uniform noise all over the image. Why the magenta cast also in the green background ? Is WB identical on both images ? Link to post Share on other sites More sharing options...
Lenshacker Posted March 11, 2015 Share #270 Posted March 11, 2015 I used LR sliders which attempt to maintain white balance, but "nothing was designed for this". Trying to put this into words... used a very long time ago for me (1980s), and seem to apply to this compression algorithm, This is a Chaotic function which increases noise level at contour boundaries and decreases it once within the step. For pixels that are on a step boundary, noise is increased. The original noise count can push the pixel into the next step. The noise will cause the pixels at step size boundaries to fluctuate between them. Once within the step, the noise gets lost in the boundary. The highlights- I want to look closer. The step size is 7 bits at that point and I am storing the difference frame. I think that the colors show up when one band is getting truncated by a large amount and the other bands fall into a step that is much closer to the real value. I am storing difference frames, only the data that gets thrown away. For the magenta cast- the same translation function is applied to all pixels regardless of color. The DNG8 algorithm is responsible for destroying the High-ISO performance of the M8. This image is using Arvid's M8RAW2DNG to get uncompressed DNG's out of the M8. iso5000eqv_s_b00_options by fiftyonepointsix, on Flickr This is ISO5000 with the M8. Most people complain about the M8 used above ISO640. The Firmware crippled it. I don't like the DNG8 algorithm because of its Chaotic behavior. I bought the M9 after buying the M8 to get uncompressed DNG. If the M8 had that choice, I would have skipped the M9 and gone straight to the Monochrom. Because it skips the Bayer pattern interpolation... "( ( Sqrt( i* 4) )**2)/ 4" <= had one open paren in previous post, but I did not upload the code, just typed this in on the fly. "Typo in pseudo-code" Link to post Share on other sites More sharing options...
CheshireCat Posted March 11, 2015 Share #271 Posted March 11, 2015 The magenta cast seems to be a "feature" of the M8 sensor. M8raw2dng – Tool to convert Leica M8 RAW files to DNG About the example ISO5000 image, I assume it has been also heavily denoised. Link to post Share on other sites More sharing options...
Lenshacker Posted March 11, 2015 Share #272 Posted March 11, 2015 The magenta cast seems to be a "feature" of the M8 sensor. M8raw2dng – Tool to convert Leica M8 RAW files to DNG About the example ISO5000 image, I assume it has been also heavily denoised. With the M8 image at ISO5000, No NR beyond just the LR defaults. From the original thread in the M8 forum: "I checked LR4.4's Noise-Reduction settings, they are the "Factory Default" : Luminosity 0; Color 25 Detail 50. I will pick a shot and disable the NR setting for a comparison." http://www.l-camera-forum.com/leica-forum/leica-m8-forum/362666-m8-raw-mode-comparison-m-monochrom.html It turns out the the M8 sensor had 3dB higher linear dynamic range compared with the M9 sensor. If the M8 had shown this performance at ISO5000 in 2006, I think some people would have been very impressed. The magenta cast- Arvid has eliminated it from v1.2 of M8RAW2DNG. Note that I am using my own software for exploring the M9 DNG files, no where near the functionality of Arvid's code. I just wanted a "visualization" of what the DNG-8 compression scheme was doing to M9 images. The picture of my daughter is with the M9. I am curious about the color cast. I have code for that. Link to post Share on other sites More sharing options...
Lenshacker Posted March 11, 2015 Share #273 Posted March 11, 2015 I scaled the differences *127 before storing back into the DNG file, then used LR sliders to bring out the image. These are from the M9. L1000441 by fiftyonepointsix, on Flickr The data being discarded by DNG8 compression, scaled for "visualization purposes only" using LR sliders. Lots of high-frequency, noise and signal. L441DFSC by fiftyonepointsix, on Flickr And a recomputation of the prior image. L439DFSC by fiftyonepointsix, on Flickr "The Daughter is in the Details". Sometimes she can be a Devil, but not here... Roy Lichtenstein comes to mind when looking at these from a few feet away. It's interesting to read some of the discussions regarding the use of DNG8 when the M9 first came out. Link to post Share on other sites More sharing options...
AlanJW Posted March 12, 2015 Share #274 Posted March 12, 2015 I think I agree with most of what your saying, it's only that I look at a lot of pictures and do see a different look most of the time between M9 and M240 Perhaps I'm wrong, but I don't think so. I'm sticking with my view I am glad you can see a difference. But I am also glad I cannot. If there are differences they are not meaningful to me, which allows me to go take pictures without agonizing about my sensor. Apparently this is true for many others. I am happy with the M240 color rendition I get in my workflow and am happy with the camera otherwise. Apparently this is true for many others also. Maybe this is "ignorance is bliss". If there are differences what we have seen is that it is hard for most to tell. Maybe not impossible but difficult and maybe impossible to say with any certainty that any differences are due to CCD or CMOS. They are different cameras made at different times in the technology life cycle, with different in camera processing going on and always improving post processing to apply. So isn't it quite amazing that the results can be so close? Link to post Share on other sites More sharing options...
Gibbo Posted March 12, 2015 Share #275 Posted March 12, 2015 Is this one of the main problems with GAS. You end up making comparisons and face the possibility of disappointment. I've only had two Leicas, an M2 and an M240 and yes, they are really different Link to post Share on other sites More sharing options...
firststream Posted March 13, 2015 Share #276 Posted March 13, 2015 David, thanks so much for all of your efforts. It's hard to believe, given the M9 M240 comparative image quality controversy that's been going on for Lord knows how long, that a simple (?!) comparison and survey such as yours wasn't done sooner. Hopefully the dust will settle, now. Kind regards. Link to post Share on other sites More sharing options...
ndevlin Posted March 14, 2015 Share #277 Posted March 14, 2015 neither Michael Reichmann nor Ming Thein has created anything that looked like their work from CCD cameras. Can they? Probably, but I guess they now like a different look for better or for worse. I encourage everyone to look at the photographs in the tests of the 645Z and tell me that they aren't processed more flat than any of the shots from the P25 in older reviews. I see Michael's files on-screen and prints in-hand regularly, and those from the 645Z match or best what earlier cameras have produced for him. And I know he certainly wouldn't go back. The amount of DR on from the Sony 50MP may make the files look more flat, but, as you say, that can certainly be dealt with in post. The video paradigm is instructive on this point, where the best raw file is a flat raw file, which allows it to be graded more later in the day. - N. Link to post Share on other sites More sharing options...
dante Posted March 17, 2015 Share #278 Posted March 17, 2015 Oh, in case you guys missed it, I posted my Part 3 conclusion yesterday. I posted the raw voting data, revealed which images were which and provided my analysis. Check it out. The Great Debate: CCD vs CMOS - Part 3 And that's the sound of Mr. Farkas dropping the mic. David, I am impressed with the survey and not surprised by the results. I'm sure some people will still be out there claiming to have the golden ear, I mean eye, but where a coin toss would do better than most survey responses, the CCD magic doesn't seem so magic anymore. Dante Link to post Share on other sites More sharing options...
CheshireCat Posted March 17, 2015 Share #279 Posted March 17, 2015 where a coin toss would do better than most survey responses, the CCD magic doesn't seem so magic anymore. Most people can't tell Bach from Mozart. Does this mean there is no difference in their music ? Link to post Share on other sites More sharing options...
pop Posted March 18, 2015 Share #280 Posted March 18, 2015 Most people can't tell Bach from Mozart.Does this mean there is no difference in their music ? Not to the people who can not hear it. Link to post Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.