Combining frequent 2-itemsets and statistical features for texture classification in wavelet domain

This paper studies a new method of texture image classification using the combination of frequent 2-itemsets and statistical features based on the discrete wavelet transform (DWT). DWT is firstly used to decompose images into different scale subbands. Then features differentiating textures for classification are extracted from these subbands. Frequently occurring local structures in images are captured from the approximation regions of one-level DWT decomposition images in the form of frequent 2-itemsets, which contain both structural and statistical information. To reduce redundancy, the paper adopts a diamond-shaped structure as the sliding window to construct transactions. Statistical features of the detail regions are then calculated and combined with frequent 2-itemsets to classify texture images. The experiments are conducted on two texture image sets, and the results show the good performance of this method.