Content-based Image Retrieval System

With the development of multimedia and Internet technique, numerous of imageinformation appears. Traditional Text-Based Image Retrieval (TBIR) cannot meet theneed of image information retrieval, so more researchers, recently, begain to focus onContent-Based Image Retrieval (CBIR).CBIR is an image retrieval technique, which synthesizes various visual feature indigital image, such as color, textual, and shapes features. Generally, the CBIR isdivided into two different sub-systems: feature extraction system and enquiry systemd.The former was mainly deviced to acquire low-level visual features and to constructimage feature vector;while the latter to make enquiry to sample image database andto retrieval object images by calculating the similarity functions.The system proposed in this paper was constructed on the base of research inrelative field including detailed ananlysis of CBIR architecture, featual extractionmethods and image match technique form a large number of reference materials andtechnique reports from within and without. In addition, some experimental resultswere provided to support our conclusion.In the retrieval process of our system, first, the image was segmented, that is, theimage is considered as a set of regions rather than pixels. And then, features ofthisregions are extracted. Here, these regional features include color, texture andshape about 9 vectors. In the last step image matching, we adopt IRM distance--amethod integrating the region and their features together to compute image differencedistancewhich is used to describe the similarity of pictures. In the image segmentationsubstep, several effective segmentation alogrithems are synthesized to improve thetime efficiency and segmentation performance. Specially, we proposed a novel imageregional weight-based algorithm, which suffer image rotation less that tridationalconterparts.This image retrieval system is robust against some ineffective imagesegmentation. In addition, it avoids the over-complexity in tridational system, andeconomizes the algorithm cost. Finally, the experiments show the effectiveness of ournew method in the CBIR system.