AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |
Back to Blog
Darktable hdr3/31/2023 ![]() ![]() Unfortunately the non-linear scaling, which is mandatory to make the integer encoding work, breaks the natural relationships between pixel values. For example, the overlay blending mode expects a middle-gray encoded at 50% (or 128 in integer encoding). Most of the image-processing algorithms used in these workflows have been tuned around these assumptions. They rely on the assumption that the image has been prepared for display at an early stage in the processing pipeline, and embed hard-coded assumptions about the RGB values of black, middle-gray and white. These workflows, using bounded RGB representations and possibly non-linear transforms to encode RGB signals, are called “display-referred”. Anything outside of this range overflows and is clipped to the nearest bound. The 8 bit integer formats are also technically limited to the 0-255 range. ![]() ![]() However, due to the use of an integer format (which implies rounding errors) they had to apply a “gamma” (essentially a transfer function applying a power 1/2.2 or 1/2.4 to encode the RGB values) and increase the bit-depth in the low-lights in order to reduce rounding errors there (humans are very sensitive to low-light details). These applications processed images encoded with 8 bit unsigned integers because it was more memory- and computationally-efficient. Most image processing applications come from the 1990s and/or inherit a 1990s workflow. ![]()
0 Comments
Read More
Leave a Reply. |