Texture Recognition Using the Wold Model and Support Vector Machines

A new method, based on the Wold texture model and support vector machines (SVMs), is proposed for texture recognition to alleviate the difficulties of charactering texture with different rotation and scale changes. First, Fourier transform and adaptive power spectrum decomposition are performed. Sector energy and ring energy of spectrum are extracted and their mean and standard deviations are calculated as texture features. Then, a texture image is rotated to place its dominant direction at 0° according to the spectral energy distribution. Co-occurrence-matrix-based features and wavelet statistical features of the rotated image are calculated as basic texture features. Texture recognition using the proposed method shows a high performance conducted with two different texture databases both of which include 25 kinds of monochromatic natural textures.