Research And Implement Of Content-based Image Retrieval System

With the growing use of digital image, how to rapidly and accurately retrieve the desired images from large-scale image database becomes a serious problem. Studies on content-based image retrieval (CBIR) technology is for the purpose of retrieving relative images effectively from image database. CBIR technology has become an active research field, and has been applied to a wide range of areas.This paper focuses mainly on the topic of CBIR technology, and systematically researches the methods of low-level image features extraction and corresponding feature matching technology, which are related to the features of color, texture and shape. The main contributions of this paper are summarized as follows:Use a quantized CIELAB space to acquire segmented color layout, and extract color feature on the basis of color histogram theory. Employ multi-resolution channel-energy model for texture analysis, and use gabor filters to extract texture features which have different orientations and scales. On the basis of the famous algorithm of straight line segments extraction, use perceptual grouping theory to group line segments into meaningful structures, such as L junctions, U junctions, parallel groups.Implement a CBIR platform based on B/S architecture, and build an image database which contains 10000 images. This system supports database image retrieval and local image retrieval and uses the features mentioned above and corresponding weight values as retrieval criteria. Perform experiments to illuminate the applicability of each feature to image categories which are composed of plants, constructions, vehicles, and natural sceneries. It proves that the method of using all three features can significantly improve retrieval accuracy.