Scalable video fingerprinting for content authenticity

Patent metadata

Field	Value
Publication number	US-12417245-B2
Application number	US-202318473045-A
Country	US
Kind code	B2
Filing date	Sep 22, 2023
Priority date	Sep 22, 2023
Publication date	Sep 16, 2025
Grant date	Sep 16, 2025

Abstract

Embodiments are disclosed for performing content authentication. A method of content authentication may include dividing a query video into a plurality of chunks. A feature vector may be generated, using a fingerprinting model, for each chunk from the plurality of chunks. Similar video chunks are identified from a trusted chunk database based on the feature vectors using a multi-chunk search policy. One or more original videos corresponding to the query video are then returned.

First claim

We claim:

1. A method comprising:
dividing a query video into a plurality of chunks;
generating, using a fingerprinting model, a feature vector for each chunk from the plurality of chunks of the query video;
identifying similar video chunks from a trusted chunk database based on the feature vectors by:
determining nearest neighbor values for the plurality of chunks of the query video and a plurality of trusted chunks corresponding to videos from the trusted chunk database, and
sorting the videos from the trusted chunk database based on the nearest neighbor values; and

returning one or more original videos corresponding to the query video based on the identified similar video chunks.

2. The method of claim 1 , wherein dividing a query video into a plurality of chunks, further comprises:
clustering the plurality of chunks into a plurality of clusters based on their corresponding feature vectors; and
assigning each cluster a cluster identifier.

3. The method of claim 2 , further comprising:
identifying a plurality of consecutive chunks associated with a same cluster identifier; and
merging the plurality of consecutive chunks into a merged chunk.

4. The method of claim 3 , further comprising:
generating a combined feature vector corresponding to the merged chunk.

5. The method of claim 4 , wherein the combined feature vector is an average of a plurality of feature vectors associated with the plurality of consecutive chunks.

6. The method of claim 1 , further comprising:
aligning the query video with an original video by computing chunk-wise distances of a plurality of consecutive chunk sequences from the original video; and
ranking the plurality of consecutive chunk sequences based on the chunk-wise distances.

7. The method of claim 1 , wherein a duration of the query video is greater than five minutes.

8. A non-transitory computer-readable medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:
dividing a query video into a plurality of chunks;
generating, using a fingerprinting model, a feature vector for each chunk from the plurality of chunks of the query video;
identifying similar video chunks from a trusted chunk database based on the feature vectors by:
determining nearest neighbor values for the plurality of chunks of the query video and a plurality of trusted chunks corresponding to videos from the trusted chunk database, and
sorting the videos from the trusted chunk database based on the nearest neighbor values; and

returning one or more original videos corresponding to the query video based on the identified similar video chunks.

9. The non-transitory computer-readable medium of claim 8 , wherein the operation of dividing a query video into a plurality of chunks, further comprises:
clustering the plurality of chunks into a plurality of clusters based on their corresponding feature vectors; and
assigning each cluster a cluster identifier.

10. The non-transitory computer-readable medium of claim 9 , wherein the operations further comprise:
identifying a plurality of consecutive chunks associated with a same cluster identifier; and
merging the plurality of consecutive chunks into a merged chunk.

11. The non-transitory computer-readable medium of claim 10 , wherein the operations further comprise:
generating a combined feature vector corresponding to the merged chunk.

12. The non-transitory computer-readable medium of claim 11 , wherein the combined feature vector is an average of a plurality of feature vectors associated with the plurality of consecutive chunks.

13. The non-transitory computer-readable medium of claim 8 , wherein the operations further comprise:
aligning the query video with an original video by computing chunk-wise distances of a plurality of consecutive chunk sequences from the original video; and
ranking the plurality of consecutive chunk sequences based on the chunk-wise distances.

14. The non-transitory computer-readable medium of claim 8 , wherein a duration of the query video is greater than five minutes.

15. A system comprising:
a memory component; and
a processing device coupled to the memory component, the processing device to perform operations comprising:
obtaining a video library comprising a plurality of digital videos;
for each digital video from the video library:
dividing the digital video into a plurality of chunks; and
generating, using a fingerprinting model, a feature vector for each chunk from the plurality of chunks of the digital video;

clustering the plurality of chunks into a plurality of clusters based on corresponding feature vectors for the plurality of chunks; and
generating a search index for the plurality of chunks obtained for the digital library based on cluster data for the plurality of clusters.

16. The system of claim 15 , wherein the operations further comprise:
assigning each cluster a cluster identifier.

17. The system of claim 16 , wherein the operations further comprise:
identifying a plurality of consecutive chunks associated with a same cluster identifier; and
merging the plurality of consecutive chunks into a merged chunk.

18. The system of claim 17 , wherein the operation of generating a search index for the plurality of chunks obtained for the digital library further comprises:
generating a combined feature vector corresponding to the merged chunk, wherein the combined feature vector is an average of a plurality of feature vectors associated with the plurality of consecutive chunks; and
generating the search index using the merged chunk and the combined feature vector.

Assignees

• Adobe Inc

Inventors

• Sinha Ritwik
• Swaminathan Viswanathan
• Jenni Simon
• Tanjim Md Mehrab
• Collomosse John

Classifications

• G06F16/75

  Clustering; Classification · CPC title

• G06F16/738 Primary

  Presentation of query results · CPC title

• G06F16/783

  using metadata automatically derived from the content · CPC title

• G06F16/7328 Primary

  Query by example, e.g. a complete video frame or video sequence (graphical querying G06F16/7335) · CPC title