Project · 2025-11-24 · 3 min read
Find in Video
Explore a video search MVP with reusable indexing, exact duplicate detection, visual fingerprints and temporal verification for matching clips.
Find in Video helps teams locate exact duplicates and partial matches across video archives. It creates a reusable index of the library, then compares clips by their visual content and sequence—not just their filenames. Users can search with a new video or select an already indexed file to find matching footage within longer recordings.
The challenge
The same footage can appear in many forms: a renamed file, a compressed export, a shorter clip, or an excerpt inside a longer recording. Exact file checks miss these variations, while manually reviewing videos becomes impractical as an archive grows.
The challenge was to identify meaningful overlaps without repeatedly processing the entire library, even as video collections grow and new exports are added.
What we built
A reusable video index
The system scans a selected folder and its subfolders, extracting compact visual fingerprints. It saves the index for subsequent searches and reuses fingerprints for unchanged files, processing new or modified videos as the library evolves.
Search from a clip or an indexed file
Users can drop a video into the interface or search using a file already in the index. Searching an indexed file reuses its stored fingerprints, avoiding another decoding pass for that query.
Exact and partial matching
File hashing provides a direct route to byte-identical duplicates. Visual fingerprints help identify matching footage when files have been re-encoded, resized, or trimmed.
Candidate lookup narrows the search before more detailed verification, reducing the need to compare every frame against the whole archive.
Verification across the timeline
A similar frame alone does not establish that two videos contain the same sequence. The matcher checks how corresponding frames align over time and looks for sustained matching segments.
This allows a short clip to be compared with a longer source recording while reducing false matches caused by isolated visual similarities.
Our role
Perfectory designed and implemented the indexing workflow, fingerprint extraction, candidate lookup, and temporal verification. We also built the browser interface for search and structured the API and processing components so the matching logic can be extended into other workflows.
A Python fingerprint sidecar supports the broader service architecture. The implementation separates the working search path from infrastructure prepared for that deployment.
Results and value
The project provides a working foundation for searching video archives by content. It supports exact duplicates and partial matches, reuses indexed fingerprints between searches, and supports repeated searches across growing video libraries.
The current delivery is an MVP. Production deployment and licensing remain separate steps requiring further integration, operational validation, and packaging.
Get in touch