Leica M 9, part 8B1 (december 21, 2009)
Leica testcharts additional info.
There seems to be some confusion and some questions about the origin of my test images. Let me explain. All my pictures with M8 and M9 and all other cameras I use, are always captured as RAW files, sometimes complemented by JPG files for easy viewing.
For the M9 resolution tests I set the M9 to uncompressed DNG and made the pictures. The files have a size of 36.4 Mb. These files are processed by the RAW programs of choice, in this case it was Lightroom 3.0 beta, as Adobe stated that this version has an improved raw engine. The resulting pictures were saved as TIFF files (typical size 54.3 Mb) and reduced to size for web publishing with GraphicConverter and converted to JPG to keep the size manageable. Some readers of my tests assumed that these reproduced JPG images were the original files and speculated that the color artefacts in the images have been caused by the original JPG algorithm. I can state that the color artefacts are visible in the original DNG files as processed by LR.
I used LR because it is recommended by Leica and because it is the most used RAW program in the market today. One could compare LR with Kodak D76 in AgX terms. It is very good, widely used and everyone understands its characteristics.
There are alternatives of course. especially in the dedicated RAW converter domain. At least a hundred such programs are available for free or a handful of dollars. Most of these programs use the raw engine “ dcraw” developed by David Coffin (dc). THis program is also used by Silverfast DR for example and also by Imatest. On the Mac we have RAW developer that uses dcraw but adds some sharpness algorithms. This program may be seen as a proxy for most dcraw-based programs as it is one of the best implementations.
I also tried Capture One, now in version 5 and Picture Window Pro, in version 5 too.
All these programs exhibited the color artefacts after the DNG file was interpreted. To avoid misunderstandings the precise protocol is this: DNG file is opened in the program and where necessary the sharpness is adjusted or when there are several algorithms available, every one is tried to see what happens. The best result is selected. The file is then exported to a TIFF file and the image is checked again to see if the original information has not been altered. Then the TIFF is opened with GraphicConverter, the small section of the file is selected and exported as TIFF again. Checks are made to ensure nothing has changed. These files are the ones you see here. The size is much bigger that the original JPG files, but there is now no possible contamination of the information by the conversion to JPG.
I can state that there is no difference between the DNG-TIFF-TIFF selection flow and the original DNG-TIFF-JPG selection flow.
There is the question what is the origin of the color artefacts: the in-camera processing to create the DNG files might be the cause or it is the post processing software. If it is the case that the post processing programs create the artefacts, then most programs produce the same results. It is typical for Leica DNG files however, because these programs when processing Canon or Olympus or Nikon files do not show the same pattern. The artefacts can be found in the very fine resolution area beyond the limiting resolution, the so-called spurious resolution. In this area the RGB pattern goes wild and that may be a cause for the color pattern.
I have found one program that interprets the Leica DNG files without color artefacts. It is a rather strange program called SharpRawPro and this one uses a neural network algorithm to convert the raw files. The base engine is again dcraw, but embedded in a neural network system. This is the only program I know that produces clean images, but here the reproduction of very fine detail is slightly less good than the best of the comparable programs. This result complicates the question: is this program smarter than the others or are the artefacts added to the DNG file by the program. More analysis is required, but one issue at least we can discard: it is not the JPG algorithm that is the villain, because that one is not part of the equation.
Note too that the images shown here are the image files as processed by the software and directly put to the web: there is no degrading by the printing process. The CaptureOne image seems to resolve the 1.25 pattern on screen and that translates into 62.5 lp/mm. The original prints by LR produced 55 lp/mm. This is a difference, but in this case we are comparing apples to pears. For comparison purposes I have processed the original DNG file with LR with the same parameters as the other programs.
Below the original DNG image processed by LR 3 beta
Below: the original DNG image processed by Capture One
Below the original DNG image processed by RAW developer
..
Below the original DNG image processed by SharpRawPro
