Classifying Texture Images Based on WP-RS and COM-RS Methods

Texture image classification is more and more important. Multi-resolution analysis methods such as wavelet and wavelet packet decompositions are more superior to other classic statistical methods. The wavelet analysis has been intensively used for texture classification with encouraging results. In this paper, two new hybrid methods for invariant pixel regions texture image classification are proposed, which are named wavelet packet rough set (WP-RS) and co-occurrence matrix rough set (COM-RS). i.e., the feature extraction and classification processing performed using WP-RS and COM-RS in this study respectively. In experiment and verification, sixty 32times32 image regions were randomly selected (overlapping or non-overlapping) from each of these twenty images of size 512times512, which obtained from Brodatz image album. Thirty 32times32 image regions and other thirty image regions are used for training and testing of the WP-RS and COM-RS methods, respectively. The experiments have run more than 100,000 times. The experimental results show that the proposed COM method outperforms the listing methods.