Skip to content

Remove Similar Thumbnails

The Remove Similar Thumbnails tool removes consecutive thumbnails that look visually similar, keeping only the first thumbnail of each "similar run." This is useful for cleaning up videos that have many near-identical frames from slow scenes, static camera angles, or long pauses.

How It Works

The tool uses perceptual hashing (dHash) to compare thumbnails. Each thumbnail is reduced to a small fingerprint that captures its visual structure. Consecutive thumbnails are compared to the last kept thumbnail -- if the similarity exceeds the threshold, the thumbnail is removed. If it is different enough, it becomes the new reference point.

This "greedy sequential" approach handles gradual transitions correctly. For example, if thumbnails A, B, and C form a slow transition where A is similar to B and B is similar to C, but A is not similar to C, then B is removed (similar to A) while C is kept (different enough from A).

Batch Processing

You can run the tool on multiple videos at once. Access it from:

  • Right-click a video and select Index > Remove Similar Thumbnails...
  • Maintenance screen using the Deduplicate button

The batch dialog provides:

  • Similarity threshold slider -- Controls how aggressive the removal is. The range is 50% to 99%. Higher values are more conservative (only near-identical thumbnails are removed). Lower values are more aggressive. The default is 90%.
  • Video selection -- Choose to process all videos, the currently selected videos, or the filtered set.
  • Progress display -- Shows per-video progress with a count of kept and removed thumbnails.

You can stop processing at any time by clicking the Stop button.

Automatic Removal During Indexing

You can also configure the tool to run automatically after indexing. Enable Remove similar thumbnails in the Video Indexer preferences tab and set the desired threshold. This runs after all other analysis (face recognition, scene classification) is complete, so keywords and face data are preserved before similar thumbnails are removed. This setting does not apply when capturing a single frame.

Tips

  • Start with the default threshold of 90% and adjust based on results.
  • A threshold of 95% removes only near-identical frames (very conservative).
  • A threshold of 70% removes frames with moderate differences (aggressive).
  • The tool processes thumbnails in frame order, so it works best when thumbnails are captured sequentially from the video.
  • Hashing is fast (~1ms per thumbnail) so the tool works well even on videos with many thumbnails.