Image Retrieval Based On Combined Features Of Interest Points

With the rapid development and increasing popularity of Internet and multimediatechnology, a large number of images are produced. In order to improve the storage,retrieval and application of expanding image databases, the image retrieval technology hasbecome a research hotspot today. In order to improve the retrieval accuracy, this paperstudied interest points and extracted color, texture and shape of interest points in the ringareas as combined features to overcome the shortcoming of a single feature. At the sametime, the feedback mechanism between users and retrieval system has been integrated withuser-friendly interface. The main works are as following:(1) SURF (Speeded Up Robust Feature) algorithm is used to detect interest points andis simplified according to the actual need. The number of points and their locations aredetermined.(2) The retrieval algorithm based on interest points is studied and a novel imageretrieval algorithm based on combined features is proposed. The right number of ring isfound and the ring idea is introduced into the Gabor wavelet texture feature to add spatialinformation. The experimental results show the effectiveness of ring idea for the texture andcombined features over single feature.(3) The feedback algorithm based on SVM (Support Vector Machines) is studied andan image retrieval method based on interest points and SVM is implemented. Theexperimental results show that this method is effective. The method uses cumulationsamples instead of single feedback samples to effectively improve the retrieval accuracy. Inorder to lighten user’s burden and improve the efficiency, memory label is used for markingretrieval result. The problems of multiple feedback are analyzed and a solution is given.These works make pavement for implementing an image retrieval system based on interestpoints and SVM.(4) A user-friendly and handy image retrieval system based on above algorithms hasbeen implemented using GUI of Matlab. This system provides the following functions:detecting and marking interest points, selecting among retrieval methods, extracting featuresand measuring similarity, marking retrieval images for implementing feedback, statistics and so on.