The Research Of Content Based Image Retrieval Algorithm

In the field of intelligent information management, image retrieval is a very important technology. It is the premise to get in-depth information from images such as object tracking,object classification, information statistics, behavior understanding and etc. As a branch of image retrieval, content-based image retrieval has wide applications and received wide attention and research. However, digital images are too many and quality is too high to get accurate results from traditional CBIR system. Therefore, many researchers put forward a lot of efforts to improve the retrieval accuracy. The purpose of this study is to improve the accuracy of CBIR with little computational cost added. The main work is as follows:Firstly, a new algorithm is put forward which is combined DCT transform and block median algorithm. The algorithm is mainly used to solve the issue that how to distinguish images similar in the frequency domain but different in color domain. The algorithm combined information of frequency domain and color domain and gets a more accurate feature vector to reflect the characteristics in color domain and frequency domain. The experimental results show that the method can improve the accuracy of image retrieval.Secondly, a new image retrieval algorithm presented combining the saliency object extraction and BOW model. The algorithm uses saliency object extraction algorithm to get main object in images and then use BOW model to compute the feature vector of the main object. At last, the algorithm uses the feature vector of main object instead of feature vector of whole image in KD-Tree search algorithm to get retrieval result. The experimental results show that the method can improve the accuracy of image retrieval that contains saliency object.