Study Of Texture Segmentation Methods Based On Wavelet Transform And Clustering Technique

Texture segmentation, as one of the key research fields in Image segmentation, plays an important rule in object recognition, image interpretation and computer vision. Feature extraction and feature classification are two key problems extensively studied in the research field of texture segmentation. By use of wavelet theory and clustering method, this paper is devoted to the study of the two key problems. (1) The texture feature extraction method based on multi-resolution and multi-channel has been studied. First of all, a texture feature extraction algorithm based on pyramid structured wavelet transform is introduced. And then, on the analysis of pyramid structured wavelet transform and discrete wavelet frame transform, an effective texture feature extraction algorithm has been proposed, using the modified discrete wavelet transform. (2) Aiming to solving the boundary effect appearing in calculating the feature values, a smoothing algorithm of texture feature based on quarter partition is put forward to improve the segmentation precision to some extent. (3) A filtering algorithm based on k-d tree is put forward to decrease the load of the k-means clustering calculation. At last, the algorithm is applied to the texture segmentation and computer simulation is performed to demonstrate the effectiveness of the presented algorithm. (4) With respected to determining the cluster number and cluster centers in using of the fuzzy c-means algorithm, the mean shift algorithm is introduced to estimate the cluster number. And then, texture segmentation with multiple resolutions is carried out from bottom to top by combining the fuzzy c-mean clustering and the peer group determined through fisher discrimination, which results in an unsupervised texture segmentation based on the mean shift. At last, computer simulation is performed.