Texture Images Recognition Based On Multiscale Geometric Analysis And Biomimetic Pattern Recognition

As a popular existent image pattern, texture becomes important research problem of computer vision and image processing. People have proposed lots of feature extraction and classification algorithms and achieved great successful in the last few years, but there are also many algorithms based on tradition mode grader, so influence validity and velocity of classification algorithms in the extent. We mainly research on two aspects is this paper: Using Contourlet transform pick-ups character vectors of texture images and exerting Biomimetic Pattern Recognition processes recognition of character vectors.Multiscale Geometric Analysis is a kind of availability semaphore and image processing method based on wavelet transform in the last few years. In this paper, educing character pick-up tool-Contourlet transform, different from wavelet transform, contourlet transform add direction gene to Multiscale Geometric Analysis. First of all done disperse field of theoretics basic of Contourlet transform, following extends sequence field. At last, to introduce kinds of applications in image processing fields. The neural network based on the biomimetic pattern recognition principles is built. The biomimetic pattern recognition makes recognition from the views of"matter cognition"instead of"matter classification", which analyzes and cognizes the high dimensional geometrical distribution that consists of the sample sets in the high dimensional feature space. It provides the theoretic basis of building the neural network bas -ed on high dimensional theory.This novel texture image retrieval system is conducted on Vistex database of 11 kinds texture images. The experim -ental results demonstrate the scheme is reasonable and efficient. First, pre-processing texture images with use of Contourlet transform, to reconstructure images with use of multi-scale energy in the direction of a significant sub-band, to describe the texture information of the different frequency bands separately, and then re-reconstructed image of the sample, calculated with the multi-resolution characteristics of the feature vector, and finally the use of ultra-sausage model of neural network identification of texture image, and the best recognition rate accomplishable 99.45%. The experimental results from various aspects of the program proved the effectiveness of the reasonable.