OCR-like recognition for poster images with text

[technopagan]

"I've tried to optimize attached image with adept-jpg-compressor, and found that it introduces visible artifacts.

I believe that this misoptimization is caused by suboptimal selection of salient region. It looks like the
SaliencyDetector is tuned for photographic images, while this image contains both photo and text over simple
background (it's a movie poster).

It would be great if SaliencyDetector could include this simple background region into salient region (e.g.,
if --poster option is passed in the command line).

Is it feasible to implement this feature?"

• Konstantin Tokarev