This brief article is a response to a discussion on the Olympus Mailing list regarding noise patterns in boosted image files from the Canon 5D Mk2. Evidently, this camera may have a pattern noise which becomes visible when the exposure is boosted during RAW conversion. I know that my old Minolta A1 has tremendous pattern noise caused by internal RFI, but what about my old Olympus E-1? The E-1 has been known, from the outset, to be a noisy camera, but it is also known for a wide dynamic range and color fidelity. What happens when you apply an extreme boost to an E-1 RAW file? 

With the exception of the final image on this screen, all images have been downsized to 800x600, sharpened, bordered and saved at JPEG 95% in Picture Window Pro.

The image shown above is the file as converted in RawTherapee with no exposure correction. This is faithful to the original RAW file as much as I can get. The grayscale boxes 14-15 is about the last tonal gradient step visible.

This next image is from RawTherapee with +5 exposure compensation. Yes, you read that right. The exposure compensation was set to +5. The way RawTherapee works, is this is applied during the RAW conversion process and written to the 48-bit file (in memory) before any additional bit-bending. No further magic was performed on this image--including noise reduction. This is as is with +5 boost.

Yes, this is noisy, but we've taken an ISO 100 image and boosted it nearly five stops. One thing that does stand out to me is how well the colors have held without looking bland. Color separation in all of the color bands, except for yellow is maintained extremely well.

What about ACR? I present the following screen shot of Adobe's ACR with the exposure controls set to Auto.

Obviously there are differences in how the RAW file converters treat the image, but you can see that Adobe applied a +3.75 exposure compensation while adding 66 brightness and increasing contrast by 37. Inotherwords, a pretty aggressive boost.

But back to the original premise of this article. What about pattern noise? The below picture is a 100% crop of the RawTherapee +5 conversion. This image is completely unaltered with absolutely no noise reduction. A worse-case scenario.

You can see in the above image that there is a randomized noise pattern which looks like it could be a pattern noise, but in a more organic fashion. This is due to the noise dithering algorithm that is employed in-camera during the analog-to-digital process. By utilizing noise dithering, the transitions at the first bit and the last bit of the readable dynamic range of each channel is randomized preventing solarization artifacts. In the case of the E-1, this dithering is most likely across the entire dynamic range and not just restricted to first and last bits which is why the E-1 has a "film-like" characteristic as well as poorer noise measurements in comparative tests. 

Ken Norton