Dual Weighted Multi-feature Texture Segmentation

Texture image segmentation is a very important research subject in imageprocessing, pattern recognition and computer vision, and it has long been paid attentionby the researchers. Texture is a very important attribute which is the same as gray leveland shape. Texture is widely used in many fields, such as in landform classification,multimedia content retrieval, crop detection, medical image analysis, defect detection,etc. Texture analysis has gradually become the core field of the image research and thecomputer vision.Texture segmentation is an important part of the texture analysis, it means that weextract the same class texture regions in a texture image containing a variety of classes,or find out the boundary between different texture. According to the different of texturefeature extraction algorithm, texture segmentation can be divided into four categories:statistics method, structure method, model method and signal processing method. Avariety of texture analysis method can effectively extract the texture characteristics, andhas been greatly developed in the practical application. But in the face of complexnatural texture images, this method usually becomes weak. In addition, due to theimperfect texture expression, and the researches on the mechanism of the humanperception of texture is not enough mature. Because of the above two reasons, whetherbefore or now, texture image segmentation is a big problem in image processing.In this paper, on the basis of the reading of vast amount of literature, we studiedthe texture image segmentation. A new texture analysis method, the combination of theGray Level Co-occurrence Matrix and the Quaternion Wavelet Transform (QWT) todescribe the image texture features. Using modified ReliefF algorithm and correlationmeasure to solve the problem of weights inside feature; using Support Vector Machine(SVM) to solve the problem of weights between features. The experiments show thatthis method has a good performance in synthetic textures and natural texture images,and has higher segmentation accuracy than other feature weighting algorithms.