Video files are actually not processed. The fingerprint is also not based on the thumbnail, but on a larger version of the image. So we would have to write an image loader for video files that loads a full frame. Currently we are fast forwarding about 10% of the video to generate a thumbnail. Whether this is also helpful for fingerprints has yet to be tested.

Maik

Oh, it seems it's a lot more complicated for videos than it is for still images. But still, if you could find a way to make this work for videos, it would be a great tool to have. Thank you for looking in to this.

I just had a thought about how this could be implemented. How about, along with the visual content of the videos, to also use the sound of the videos to try to match them? Kinda like how 'Fuzzy View' works for still images, but for the sound. Some kind of algorithm like what YouTube has for finding if someone used a song in their videos. It can also be customizable about how accurate the user would want it to be. YouTube's algorithm can find a song even if someone just whistled a tune. I am not a programmer but I think the waveform of the sound of the videos should be unique enough to be accurately fingerprinted.

I can imaging a scenario where someone edited a wedding video and wants to see how many small clips he used on the main, longer and final video. Using time stamps would not work in this case because the length of the videos doesn't match. But if you look for a waveform fingerprint of the smaller clips if they are inside the waveform of the final and longer video, then this should give the desirable results.

Of course this would not work for videos without sound, so this should not be the only think that is checked. Also, when the fingerprints of the videos are made, the function should ignore sections that happen to be silent, it should choose a spot in the video with sound enough to make a good waveform fingerprint.