Configurable sample location

[whacklezz]

Describe the solution you'd like Being able to choose a different point in time of the video to pick the sample that is used for comparison would allow for more flexibility in use.

I have done this manually by editing the source code and picking something else than the default (n/n+1)*length slicing (n being the number of screenshots to compare). It has allowed me to find a lot of additional duplicates by sampling a single screen e.g. at 0.1 or 0.8*length.

If this could be configurable in the program, it would be awesome.

Thanks for making a really nice program, keep up the great work!

[Maltragor]

I had also experimented with a modified duplicate search with a similar objective some time ago. However, my approach was based on being able to specify that a configurable number of bad matches be ignored. The position can still only be determined indirectly via the number of thumbnails, but since not every thumbnail is taken into account, it is comparable to setting different individual positions one after the other.

• There are still n thumbnails created in the previous grid
• You set a number m of worst matches to be ignored
• When comparing the thumbnails in pairs, the result is added into a list. This list is sorted according to the similarity value after all n comparisons have been made. The m worst matches are ignored, i.e. they are neither included in the mean value nor are they compared with the limit.

The application scenario and the idea was: If actually similar videos have a slight time offset or possibly already due to different locations of the key frames when compressing, a thumbnail position can coincide with a scene change. In one video the thumbnail is then created before the change, in the other afterwards so that they are completely different. If you create a few more thumbnails than necessary and ignore the worst matching pairs, such videos will still be recognized.