Fusion of Rotation-Invariant Texture Features for

Multichannel Gabor filters (MGFs) and Markov random fields (MRFs) are two common methods for texture classification. However, the two above methods make the implicit assumption that textures are acquired in the same viewpoint, which is unsuitable for rotation-invariant texture classification. In this paper, rotation-invariant (RI) texture features are developed based on MGF and MRF. A novel algorithm using the neighborhood-oscillating tabu search (NOTS) is proposed to fuse RI MGF and MRF features, compared with the sequential forward floating selection method. Experimental results indicate that the fused RI MGF/MRF features achieved by NOTS have much higher discrimination than pure features in terms of classification accuracy.