Abstract

Content Based Image Retrieval (CBIR) becomes possible and necessary as computer graphics, machine learning, knowledgebase and database technologies mature. SHMM aims to be a web oriented image library with an automatic image addition and classification mechanism to support sample image based similarity search, and semantic description search as well as allowing the registered users to add image to the database. As a sample CBIR system, SHMMhas its feature extraction layer, which supports colour and texture feature extraction that generates a 16-dimensional vector value. Based on this vector, a 16-dimensional SR tree is constructed. Using the nearest neighbor search technology on the SR tree, similarity search is supported. With backend database support, semantic description search, which is based on the keyword of the semantic meaning of image, is also implemented in SHMM. When a new image is added to the system, SHMM will automatically scan its feature, suggest the semantic description based on the similarity search result in the library, and wait for the user's response. Image contributor can accept the system's suggestion or inform system administrator via email to create new semantic description category in the system. (Abstract shortened by UMI.)

Document Downloads