Efficient Feature Extraction for Robust Image Classification and Retrieval

In this paper, a new feature extraction method for robust image classification and retrieval is proposed. The robust image classification and retrieval systems are required when the images are not ideal such as geometrically distorted and/or contain additive noise. To construct an efficient feature space, an optimum linear transform is obtained by nonlinear optimization in learning process using a set of image samples. In the simulations, the method is experimentally applied to characterize wavelet packet representation of texture images robust to noise and geometrical (rotation and translation) distortion. Further, it is efficiently used for texture retrieval system to demonstrate the usefulness of the method. It is shown that the higher retrieval rate is achieved compared with the conventional approach such as discriminant analysis