support perceptual hashing

[dave rice]

Summary of the enhancement:

I would love to see FFmpeg support for perceptual hashing per frame (either as a filter output or format output). The results could be stored in order to enable perceptual search across collections of file of frames.

[Carl Eugen Hoyos]

What is perceptual hashing?

[dave rice]

Here's a good introduction to it: ​http://bertolami.com/index.php?engine=blog&content=posts&detail=perceptual-hashing. Also see the citations at ​https://en.wikipedia.org/wiki/Perceptual_hashing. Some image search cites that use perceptual hashing are ​http://tineye.com/ and ​http://images.google.com/.

A use case could be:

• a user has a set of 1,000 files representing source material and finished work encoded with various lossy encodings
• something like ffmpeg -i sourcefile.mkv -f perceptualhash sourcefile_perceptual.txt could be generate per file which would contain something similar to the output of framemd5 but with perceptual hashes instead of md5 values
• the perceptual hashes could then be indexed to reveal relationships between the media. i.e. these source files contain frames that look like frames in this edited work.

[dave rice]

related to this thread: ​http://ffmpeg.org/pipermail/ffmpeg-devel/2016-March/191750.html

[dave rice]

Resolved by ​https://git.ffmpeg.org/gitweb/ffmpeg.git/commit/5e3a418b6047acd848698c4bb4bf0c1b73526744