Leica M9 Raw Comparison
Leica M9 RAW Comparison
We've often look at RAW files converted with dcraw, an excellent freeware raw converter. dcraw usually offers timely support for the latest cameras, but more importantly, it does not apply any noise-reduction, sharpening or other corrections such as geometric distortion correction to the output files. There will always be differences between RAW converters, in terms of the sort of demosaicing algorithms they use (the processes by which they convert the separate Red, Green, and Blue data sets to an array of full-color RGB pixels), but dcraw seems to use a fairly generic algorithm that delivers good sharpness with relatively few artifacts, and can be counted on to not apply any noise reduction if you don't want it to.
See the crops below to compare the M9's RAW image quality to the Canon 5D Mark II, Nikon D3X, and Sony NEX-7. We would have liked to include the Sigma SD1 here, however dcraw (and Adobe Camera Raw) did not support the SD1's RAW files at time of writing.
Here, we can see the Leica M9 produces excellent sharpness in its RAW files thanks to the lack of an AA filter, however as expected it is more prone to moiré and other demosaicing errors (though at higher ISOs, chroma noise interferes with demosaicing on all four cameras). The M9's 18-megapixel CCD doesn't do as well as the CMOS sensors in the other full-frame bodies from Canon and Nikon in terms of noise, particularly in the red channel, despite its larger 6.8µm pixel pitch. (The 21-megapixel 5D Mark II's pixel pitch is 6.4µm, while the 24.4-megapixel D3X's is 5.9µm.) It also looks as though the M9 is applying some subtle noise reduction to its RAW files, which is something we'd rather not see. The M9 compares well to the 24-megapixel Sony NEX-7 with its much smaller 3.9µm pixels, though it's a little difficult to compare because of the M9's noise reduction.
Note that the Canon and Sony offer a much wider range of ISOs than shown. The 5D Mark II tops out at ISO 25,600 while the NEX-7's maximum ISO is 16,000 equivalent.