If you are getting anything like a good picture with the right number of pixels and colours, then it's likely that the media controller graph is correct.It starts to get even wilder...
When I insert a full green test pattern with a defined pixel value like 2000 (aka 125 in 8bit) for green1 and 4000 (aka 250 in 8bit) for green 2. I get peeks at 149, 154 and 157. And a standard bar chart does not show black at all and the peak are non-linear shifted. I added the pictures and histograms at the end of the post.
Can you explain more clearly what you are doing -- You are generating these patterns using the image sensor's test modes?
Were those histograms derived from the raw buffer from C++; or the raw DNG image from rpicam-still; or the final PNG from rpicam-still?
How does the unprocessed (without demosaic) raw image from C++ compare with the contents of the DNG image from rpicam-still? (Note that the DNG will always be unpacked to 16 bits per pixel). Can you get histograms for those?
The final output from rpicam-still has been processed in various ways, some of them non-linear (though it shouldn't be raising the black level...)
Statistics: Posted by njh — Fri May 10, 2024 10:19 am