using a basic perceptual hash algorithm,
trying to find a frame from a pool of videos
- video server hosting the video(s) must support ranged requests
- video server should support CORS, otherwise one has to download the file or use a proxy so the parsing step is possible
- once the video parsing ends, copy the console.log line into
results/all.js - the save screenshot button saves frame so you can look for it in the search page O:)
you should parse the video locally or use a video URL with CORS, then fire it here:
http://josepedrodias.github.io/video-hash/parseVideo.html
you can serve videos from the videos dir. just run the server.js file ([sudo] npm install at first use)
With the test results on the repos, if you submit a frame from one of these videos:
- http://rd3.videos.sapo.pt/i0Az39C8OSy6vO0rEtbk/mov/1
- http://rd3.videos.sapo.pt/uWMr3C8FQZE5FWwCLZTc/mov/1
you should get a close hit, hopefully.
Try here:
http://josepedrodias.github.io/video-hash/searchVideo.html
article http://www.hackerfactor.com/blog/?/archives/432-Looks-Like-It.html
basic pHash impl ??
scene detection http://gmarty.github.io/SCD.js/
rgb2lab https://github.com/gka/chroma.js/blob/master/chroma.js
?rgb2lab 2? http://stackoverflow.com/questions/13405956/convert-an-image-rgb-lab-with-python
histogram me
dct https://github.com/vail-systems/node-dct/blob/master/src/dct.js
- IMPROVE PHASH PRECISION 1
replace media for DCT
- IMPROVE PHASH PRECISION 2
improve pHash, by converting it to a better color space than RGB (CIELAB?) and computing 3 hashs (one for each channel). hash is the concatenation of 3 binary hashes
- COMPUTE HASH FOR THE MOVIE
sample n frames from the video (ex: 10). for each, compute the histogram (an image itself) and do a pHash of it! use those entries to look for the movie (instead of all the frames sampled)
- SMARTER HISTOGRAM SAMPLING
make use of a scene detection algorithm to identify scenes. take one histogram at every scene half