Large-Scale Partial-Duplicate Image Retrieval and Its Applications
Abstract
The explosive growth of Internet Media (partial-duplicate/similar images, 3D objects, 3D models, etc.) sheds bright light on many promising applications in forensics, surveillance, 3D animation, mobile visual search, and 3D model/object search. Compared with the general images, partial-duplicate images have some intrinsic properties such as high repeatability of local features, consistent local patch appearance, and stable spatial configuration. Compared with the general 2D objects, 3D models/objects consist of 3D data information (typically a list of vertices and faces) to represent 3D objects. However, these unique properties of partial-duplicate images and 3D models have not been well exploited to design effective and efficient search algorithms. Because of this, existing works for large-scale partial-duplicate image retrieval and 3D model retrieval suffer from two major problems: low accuracy and low efficiency. These problems make them fall far below many applications requirement. This project has investigated many key problems in large-scale partial-duplicate/similar image and 3D model retrieval: feature descriptor problem, image representation problem, index strategy problem, feature quantization problem, image search results quality assessment problem, image search reranking problem, sketch-based 3D model retrieval problem, and related search problems and has proposed a series of effective and efficient approaches to solve them.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 23, 2016
- Accession Number
- AD1011038
Entities
People
- Qi Tian
- Yijuan Lu
Organizations
- Texas State University