Some Problems On Content-Based Image Retrieval

With the rapidly development of computer network, the popularity of mobile phones and digital cameras, the amount of digital images is greatly increased, also produced a variety of databases. How to accurately and fast search the needed images in these databases has gradually attracted people’s attention. In 1990 s, people proposed the content-based image retrieval technique has become a hot research topic to solve this problem.Content-based image retrieval mainly including feature extraction and similarity matching. However, the semantic gap between low-level visual feature and high-level semantic is an intensive and difficult problem in content-based image retrieval. Relevance feedback is an effective strategy to narrow down the semantic gap, the relevance feedback based on swarm intelligence optimization can effectively find the query vector in the feedback stage, thus reduces the number of queries, saving the user waiting time.This paper mainly includes the following two aspects:One is to explore the single features and different combination of features on the effect of initial retrieval results, after that, image feature directly affects the initial retrieval results, this paper present color histogram, color histogram and color moment, color histogram and Gabor features, color histogram, color moment and Gabor features, these combined features on the effect of initial retrieval results. With group feature, find the feature group in which weight combination is optimal effect.The second is to explore how the number of iterations in Particle Swarm Optimization(PSO) affect retrieval results. The PSO algorithm is feasible in relevance feedback, and the number of iterations in the PSO algorithm is a very important parameter, therefor, this paper discusses the PSO internal parameters,the number of iterations affect retrieval results, given the results, analysis it and make the conclusion.