DTCC And Its Applicatinns In Image Mining

This thesis mainly studies the directional multiresolution transform DTCC and its application in image processing and image data mining.The research of directional multiresolution transforms is one of the basic research directions of high dimension signal processing. DTCWT, Contourlet, NSCT and PDTDFB are the newly developed directional multiresolution transforms and are widely used in kinds of image processing, analysis and mining applications. They are trying to provide the multiresolution and directional representation of high dimension signals. However, considering the characteristics such as directional selectivity, shift invariance, redundancy, phase information and computational efficiency, they all suffer from problems caused by their own structures.Image mining extracts all kinds of knowledge, identifies the relationship and reveals the hidden patterns in and among images. It is an important development and branch of data mining studies.The research work can mainly be concluded as follows:First of all, we construct a dual tree LP transform based on the motivation of DTCWT for the purpose of improving the shift invariance of LP, and combine the DFBs which ensures the higher directional selectivity, then implement a near shift invariance and higher direction selectivity transform which is named as dual tree complex Contourlet, DTCC. DTCC is a dual tree structure and inheriting the higher direction selectivity from DFBs and can be regarded as two Contourlet transforms in parallel. It has8/3redundancy at most and has phase information.Secondly, we apply DTCC to image denoising and texture retrieval. The experimental results show that DTCC is a good directional multiresolution analysis in these applications.Thirdly, we built the HMT model based on DTCC in order to reveal the dependence existing among the DTCC coefficients. Then we employ DTCC HMT in image denoising and texture retrieval, and experimental results show that the DTCC HMT outperforms the Contourlet HMT.Fourthly, we propose an image classification method based on association rule under DTCC. The transactions are constructed to describe the micro structures of a local region in the DTCC subbands, and then association rules are generated and their statistical parameters are employed as image features used for image classification. The experimental results show that the association rule mining algorithm based on DTCC can reveal the hidden co-occurrence patterns of the micro structure among images and lead to higher correct classification rate.Finally, we introduce a mining algorithm on palmprint image according to the characteristics of palmprint images and use it in palmprint image classification. A pre-processing is applied to original palmprint images, then DTCC is used to decompose them into subbands, and the association rule algorithm described before is adopted for mining. The experiments show the efficiency of the algorithm.