A Long-Term Learning Algorithm in CBIR Based on Log-Analyzing

Nowadays, the amount of images increase drastically. Content-based image retrieval (CBIR) has been proposed to efficiently manage these images. Traditional CBIR system extracts the low features of image automatically. Because of the difference between human comprehension and machine, the search results provided by CBIR system always can not satisfy the user's need. So relevance feedback has been used in content-based image retrieval (CBIR) to bridge the semantic gap, which is existing between image low-level features and high-level human perceptions. This paper proposes an extended-judging algorithm to analyze the information which includes both positive relevance and negative relevance. It uses the feedback log data and index table to expand the set of relevant images, and judging images in database by the current feedback record. Results show that compared with the traditional long-term learning method, the retrieval performance can be improved apparently.