Supervised Texture Classification Using a Novel Compression-Based Similarity Measure Mehrdad J. Gangeh 1 , Ali Ghodsi 2 , and Mohamed S. Kamel 1 1 Center for Pattern Analysis and Machine Intelligence, Department of Electrical and Computer Engineering, University of Waterloo, Ontario N2L 3G1, Canada, {mgangeh,mkamel}@pami.uwaterloo.ca 2 Department of Statistics and Actuarial Science, University of Waterloo, Ontario N2L 3G1, Canada, aghodsib@uwaterloo.ca Abstract. Supervised pixel-based texture classification is usually per- formed in the feature space. We propose to perform this task in (dis)simil- arity space by introducing a new compression-based (dis)similarity mea- sure. The proposed measure utilizes two dimensional MPEG-1 encoder, which takes into consideration the spatial locality and connectivity of pix- els in the images. The proposed formulation has been carefully designed based on MPEG encoder functionality. To this end, by design, it solely uses P-frame coding to find the (dis)similarity among patches/images. We show that the proposed measure works properly on both small and large patch sizes. Experimental results show that the proposed approach significantly improves the performance of supervised pixel-based tex- ture classification on Brodatz and outdoor images compared to other compression-based dissimilarity measures as well as approaches performed in feature space. It also improves the computation speed by about 40% compared to its rivals. 1 Introduction Texture images can be divided to two broad types: stationary that contains only one texture type per image and nonstationary that consists of more than one texture type per image [1]. The main application domain on stationary texture images is supervised classification of each texture image into one class; whereas on nonstationary texture images, there are two main application domains [1,2]. First, unsupervised texture segmentation that partitions the texture image into disjoint regions of uniform texture. Second, pixel-based texture classification, which is similar to texture segmentation in the sense that the given texture image is segmented to uniform texture regions. The difference, however, is that in pixel classification, the segmentation is performed using supervised techniques [2]. In this paper, our focus is on supervised pixel classification and hence, we deal with nonstationary texture types. Common trend in literature on pixel-based texture classification is the com- putation of some texture features for every pixel using its neighboring pixels arXiv:1207.3071v1 [cs.CV] 12 Jul 2012