Texture Analysis And Svm-based Multi-scale Image Classification And Retrieval

With the explosive growth in the volume of digital image databases, how to extract the visual information has recently become a very active research area. Effective image classification and retrieval is the key problem of obtaining image information. This paper considers the image classification and retrieval as two main processes: image feature extraction and feature's classification and retrieval using learning machine.Multiscale geometric analysis is an effective signal and image processing method developed from wavelet transform, including ridgelet analysis, curvelet analysis, contourlet transform and so on. Different from wavelet analysis, Multiscale geometric analysis imports direction factor, which makes the two dimensional presentation sparser, so it is used widely in image compression, feature extraction, modern recognition, image denoising and so on. This paper discusses the effect of feature extraction using wavelet transform and contourlet transform.Support vector machine is a general learning machine, and is a method of statistical learning theory, its main idea is structural risk minimization. In order to solve the linear nonseparable issue in the primal space, mapping the input vector to high dimensional space, and constructing the optimal hyperplane in this space. It is widely used these years because of its perfect performance. This paper uses the SVM as a learning machine for classification and retrieval.The main work includes:Palmprint identification has been developed for security purpose. In this paper, a novel scheme of palmprint identification is proposed. We apply 2-dimensional 2-band (Discrete Wavelet Transform) and 3-band wavelet decomposition to get the approximate and detailed subband images, and then use them as identification feature vectors. We choose support vector machines as classifier. The experimental results demonstrate that it is a simple and accurate identification strategy and the correct recognition rate is high up to 100%. An image retrieval scheme based on multiscale analysis and SVM relevance feedback is proposed. Firstly, more accurate texture feature can be extracted in contourlet domain than Wavelet due to its multi-resolution and directionality. One class and binary class SVM are combined to retrieve. The one class SVM can estimate the distribution of data in high dimensional space, can exploit unlabeled data to get a primary similarity measure order. Then binary class SVM is used to get the labeled sample information through learning user's feedback and finally improve the retrieval accuracy. This novel texture image retrieval system is conducted on VisTex database of 640 texture images and the average retrieval rates are up to more than 95%. The experimental results demonstrate the scheme is reasonable and efficient, the most appropriate feedback image number and feedback times are proposed.