Image Search with Relative Attribute Feedback
Computing & Wireless : Computing Methods
Available for licensing
- Kristen Grauman , Computer Science
- Adriana Kovashka , University of Texas at Austin
Current image searches use methods that learn monolithic attribute predictors, with the assumption that a single model is sufficient to reflect human understanding of a visual attribute. However, in reality, humans vary in how they perceive the association between a named property and image content, which can make relative, accurate searches difficult.
This invention proposes a user-specific attribute model that adapts a generic model trained with annotations from multiple users, tailoring it to satisfy user-specific labels. Furthermore, this offers novel techniques to infer user-specific labels based on transitivity and contradictions in the user’s search history. It demonstrates that adapted attributes improve accuracy over both existing monolithic models as well as models that learn from scratch with use-specific data alone. In addition, this innovation shows how adapted attributes are useful to personalize image search, whether with binary or relative attributes.
- Quicker search for desired content
- Search by use of a mental model
- Unique concept of relative feedback
- Allows users to express exactly what about the image is relevant
- Builds a precise and efficient searches
Image search, relative feedback