Research And Implementation Of Image Retrieval Algorithm Based On Contourlet

With the rapid development of the multimedia and the Internet technology, images are more and more extensively applied. Traditional text keyword-based retrieval is not to meet image retrieval, making content-based image retrieval (CBIR) become a hotspot owe to its practical significance. The Contourlet transform, flexibly and effectively combining the multi-scale with the multi-directionality of image representation, can describe an image accurately and optimally. Therefore, theories and applications based on Contourlet transform have great prospects in the fields of image retrieval.In this thesis, on the foundation of analyzing the key techniques of CBIR, we mainly study the multi-scale geometric analysis, Contourlet transform and Nonsubsampled Contourlet transform (NSCT), in CBIR. The main contents of the thesis are summarized as follows:1. Introducing some key techniques of CBIR systematically, such as low-level feature descriptions including color, shape and texture, the similarity measure, relevance feedback technologies, and performance evaluation.2. After analyzing disadvantages of wavelet transform in dealing with image, introducing the fundamental theories and the realization of Contourlet transform and NSCT, and making theoretical comparisons of the two transforms.3. Proposing two image retrieval algorithms based on texture and shape. Ones texture features are extracted by Contourlet transform, and sharp features by Gaussian descriptors. Another's texture features are extracted by NSCT and sharp features by Gaussian descriptors. Euclidean distance is adopted to measure similarity; The Adaboost algorithm is employed for relevance feedback.4. Comparing three algorithms based on Gabor, Contourlet and NSCT respectively in retrieval validity in the applications of image retrieval for patent designs. Experimental results indicate that the algorithm based on Contourlet and Gaussian descriptors is the most feasible and effective. Relevance feedback by means of Adaboost can highly improve the accuracy of image retrieval.