Texture Image Segmentation Based On A Synthetic Approach

As an important aspect of digital image processing and pattern recognition, texture image segmentation has always been one of the hottest and most difficult study topics. Texture image segmentation is of great significance not only to many researches of computer vision and image processing but also to practical application. Textured image segmentation comes down to all kinds of images and appears in all fields of image processing. There are a lot of applications such as cancer cell recognition in medical images and object identification in remote sense images.Texture image segmentation divides an image with different kinds of texture into several consistent texture regions corresponding to the entities in reality. It is a very difficult problem in texture image segmentation because it is hard to get information about texture patterns and species without the prior knowledge. Texture image segmentation consists of feature extraction and segmentation approaches. According to different feature-extracting approaches, there are three approaches: statistics approach, structure approach and spatial/frequency approach. Spatial/frequency approach is a new approach developed in the last decade. And it is more and more popular for its many advantages. Texture image segmentation gives a class tag to each pixel in the image.In this paper, we made an investigation into texture feature extraction based on Spectrum, clustering and classification, and a modified approach based on Gabor wavelet transformation, GMM clustering and LS-SVM classification has been provided for texture segmentation. The modified approach uses multi-channel Gabor filter to get texture feature vectors, and optimizes them, and then uses GMM clustering to transform these vectors into training samples for LS-SVM, and uses LS-SVM to complete the segmentation. Trough experiments, the modified approach can work well with problems which trouble only clustering segmentation approaches and classification segmentation approaches, so it shows that the modified approach is a texture image segmentation approach with good performance. Meanwhile, the modified approach is used to deal with stone image recognition and digital document image segmentation and runs well on them.