Abstract
User-provided titles and descriptions of video clips may lack attribution to the original source(s), making it difficult to access the original content, verify authenticity, or evaluate fair and permissible use. This disclosure describes techniques that enable identifying the original source(s) for derivative video content by comparing video fingerprints and embeddings that are derived using one or more suitably trained machine learning models that analyze the audio and video content of a video clip being evaluated. Scores are assigned to matching videos based on the correlations of fingerprints and embeddings. Matching videos with scores that meet a threshold can be shown to users as the likely original source(s) as a list of linked video thumbnails, a list of linked titles, etc. The scores can be indicated to users, e.g., a composite score for each matching video, or a breakdown of individual scores within the composite score. Time signatures of matching video segments can also be displayed. The described techniques can identify original video sources (root videos) for different types of derivative videos, such as subsets, compilations, mixes, parodies, imitations, etc.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Klein, Daniel and McCartney, Paul, "Identifying Source Videos for Video Clips Based on Video Fingerprints and Embeddings", Technical Disclosure Commons, (March 06, 2024)
https://www.tdcommons.org/dpubs_series/6752