Study Of Image Feature Extraction And Texture Classification Algorithm

Texture recognition is an important topic in the field of computer vision. Based on the traditional texture analysis methods, statistical geometrical features (SGF), gray gradient changement (GGCM) and log-polar transform are especially studied in the paper, and the texture classification system is constructed by combining these methods with support vector machines(SVM).In this paper, a rotation invariant image preprocessing algorithm based on log-polar transformation is proposed. The log-polar image of similar texture with different scale and orientation only has shift rows. The experiment results show that log-polar glcm can get higher accurate classification rate than tradional gray Co-occurrence Matrix (GLCM). In addition, GGCM is studied by adding the gradient information into co-occurrence, which can include more arrangement information of image elements.SVM has a solid theory foundation and very well classification capability. In this paper, multiple classifier based on SVM was studied to built the raw classification system and subclass system, which can achieve highly accurate texture classification . With the brodatz texture library and rotate one respectively, we can get the accurate classification rate of 92% and 99% in simulation experiments.As a new feature extraction method, SGF describes the texture with image function. The gray image is split into a series of binary images with variable thresholds. The texture description feature can be deduced by the connected domain and geometric topology property of binary images. Texture classification experiments result shows that comparing with glcm and discrete wavelet transform, SGF has very strong ability in texture description and rotation overcoming, and stronger texture recognition ability as well.