Compressed sensing for robust texture classification

Abstract. This paper presents a simple, novel, yet very powerful ap-proach for texture classication based on compressed sensing. At the feature extraction stage, a small set of random features is extracted from local image patches. The random features are embedded into a bag-of-words model to perform texture classication, thus learning and classi cation are carried out in the compressed domain. The proposed unconventional random feature extraction is simple, yet by leveraging the sparse nature of texture images, our approach outperforms tradi-tional feature extraction methods which involve careful design and com-plex steps. We report extensive experiments comparing the proposed method to the state-of-the-art in texture classication on four databases: CUReT, Brodatz, UIUC and KTH-TIPS. Our approach leads to signif-icant improvements in classication accuracy and reductions in feature dimensionality, exceeding the best reported results on CUReT, Brodatz and KTH-TIPS.