Research On Image Retrieval System Based On Local Features

With the rapid development of the Internet technology, the multimedia technology as well as the prevalence of large memory capacity electronics, the multimedia image data have been growing at a remarkable rate. How to retrieval large image database efficiently is becoming a very important and challenging research topic. Textual-based image retrieval is less efficient and the speed of search is slowly which cannot meet the demand of people and content-based image retrieval (CBIR) is expected to address this issue.Based on the analyzing and discussing for the key techniques of CBIR, this paper mainly researched on the algorithms of feature extraction and relevance feedback. The main content of this paper is summarized as follows:1. This paper proposed an improved method combining Speeded Up Robust Features (SURF). with Maximally Stable Extremal Regions (MSER). By combining SURF features into groups and measuring the geometric similarity among features, the discriminative power of the grouped features has been significantly increased. Simulations show that the proposed method outperforms the original SURF algorithm both in match ratio and repeatability.2. This paper proposed an improved genetic algorithm-particle swarm optimization (GA-PSO) algorithm and combined it with relevant feedback process effectively. The improved GA-PSO algorithm avoids obtaining local solution with applying PSO and Euclidian data distance to mutation procedure on GA’s differentiation. The method embeds the user’s query intent into feedback process, then using the proposed GA-PSO algorithm to adjust the weights for feature extraction that the user can; accept. Experiments proved that the algorithm improves retrieval capability and has achieved better retrieval result.3. This paper designed a CBIR system based on local feature and relevance feedback as the test platform and the development environment is C++. The proposed system cited the Exact Euclidean Locality-Sensitive Hashing (E2LSH) to speed up the query process. The experimental results show that the recall and precision of this proposed approach is better.Researching on CBIR requires a wide-range of knowledge. This paper has just discussed two of the key techniques in detail and more research will be done in the future.