Sparse Variation Pattern for Texture Classification

We present Sparse Variation Pattern (SVP) to extract image features for texture classification. Using the directional derivatives in a local circular neighborhood, SVP captures texture transition patterns in the spatial domain. Unlike conventional feature extraction methods, SVP characterizes the image points taking the co-occurrence of two derivatives in the same direction into account without encoding to binary patterns. Using the directional derivatives, SVP defines a dictionary to solve the classification problem with sparse representation. The proposed texture descriptor was evaluated on the FERET and the LFW face databases, and the PolyU palmprint database. Comparisons with the existing state-of-the-art methods demonstrate that the SVP achieves the overall best performance on all three databases.