Web infrastructure
ExistingDistributed crawling, collection, processing and indexing: the media discovery foundation is already built.
01Recognition technology
Identifying copies of images and videos.
Our image and video research focuses on recognising altered copies while distinguishing them from media that only look similar.
02Images / specialised research
When an image changes appearance
Simple perceptual fingerprints summarise an image’s appearance. Useful for some comparisons, they can struggle with substantial cropping or several combined alterations.
ShieldHer is developing specialised copy detection to better handle these situations. The research focuses on both resilience to alterations and the distinction between actual copies and similar content.
- Cropping, resizing and recompression
- Screenshots, borders and watermarks
- Combined visual alterations
SH / 02Copies of the same content
Source imageSame content, different appearance
Recognise the identity of the content, beyond visual similarity.03Video / experimentation
Tracing an excerpt to its source
A redistributed video may contain only a few seconds of the original production. Our work explores recognition of partial sequences that have been compressed, cropped, slowed down, sped up or degraded.
Video recognition tests have produced encouraging results on heavily altered short excerpts.
Reference sequenceA — F
A
B
C
D
E
F
Cropped excerptC — E
C
D
E
The excerpt preserves three moments of the sequence within a tighter crop.
SHFingerprints & representations
Data retained for recognition
The recognition index uses fingerprints (hashes) and digital representations. Media is analysed to produce these elements; the image and video files themselves are not retained in the index.
This enables content comparison while limiting retained data, as part of a data-minimisation approach informed by the GDPR.
Our approach to privacy04Development status
Capabilities and maturity
The closed beta builds on the indexing infrastructure. Image research and video experiments have separate scopes.
Image recognition
Specialised researchDeveloping alteration-resilient copy detection, with particular attention to false positives.
Video recognition
Experiments completedEncouraging results on altered short content. Extending this to long videos and a complete protection workflow remains a goal.
The video results come from specific experiments; they do not measure the performance of a production service for monitoring long videos. Access to the closed beta is discussed through professional enquiries.
SHCONTACT
Let’s talk about your content.
Tell us about your catalogue and the redistribution that concerns you so we can define a search scope.