imgdupes VS videohash

Hi all, I try to create a program to detect if a certain video scene (normally within 10 seconds) is within a longer video file. The idea is that if I find an scene on youtube, I want to know from which episodes of a particular TV show (assuming I know which tv show, but no idea which episode), so I want to find it out. Current solution: [a] - Extract Frame using ffmpeg from the reference clip (fps = 1) [b] - Extract Frame using ffmpeg from the longer video file (fps around 0.1 or 0.5) For each frame from [a] , I do a imagehash for [a] and [b] and comparing the hamming distance, get the lowest distance from this round of comparision and move on to the next frame from [a] Eventually I got an average score and I can find out if this TV episode contain the scene I was looking for. However, this is slow and not efficient. I found out that there is a videohash library https://github.com/akamhy/videohash But it said "Videohash cannot be used to verify whether one video is a part of another (video fingerprinting)." Does anybody know why? Is it because it gets a videohash for the whole video? If this is the case, how about I use the video hash lib to create a hash for my reference clip (let's say it is about 10 seconds) and then I create multiple 10-second version of the Longer video, generate a videohash just for it and compared that with my reference clip. Would that work? (Yes I understand that for a 60 minutes movie, that would be like 360 video hash to be calculated)... Do you think this is better? Thanks.

2) Sorting through all the scraped data. This is going to vary depending on your approach. If you want to remove duplicates, there is a library called videohash that will allow you to take the perceptual hash of video files. If 2 videos have the same perceptual hash, they are duplicate.

I think it creates a collage of the video frames: https://github.com/akamhy/videohash/blob/8759b6ad7fdabcdf4dd...

and passes that on to the videohash.py module to generate a hash: