Robust Texture Image Representation by Scale Selective Local Binary Patterns
Zhenhua Guo,Xingzheng Wang,Jie Zhou,Jane You
DOI: https://doi.org/10.1109/tip.2015.2507408
IF: 10.6
2015-01-01
IEEE Transactions on Image Processing
Abstract:Local binary pattern (LBP) has successfully been used in computer vision and pattern recognition applications, such as texture recognition. It could effectively address grayscale and rotation variation. However, it failed to get desirable performance for texture classification with scale transformation. In this paper, a new method based on dominant LBP in scale space is proposed to address scale variation for texture classification. First, a scale space of a texture image is derived by a Gaussian filter. Then, a histogram of pre-learned dominant LBPs is built for each image in the scale space. Finally, for each pattern, the maximal frequency among different scales is considered as the scale invariant feature. Extensive experiments on five public texture databases (University of Illinois at Urbana-Champaign, Columbia Utrecht Database, Kungliga Tekniska Högskolan-Textures under varying Illumination, Pose and Scale, University of Maryland, and Amsterdam Library of Textures) validate the efficiency of the proposed feature extraction scheme. Coupled with the nearest subspace classifier, the proposed method could yield competitive results, which are 99.36%, 99.51%, 99.39%, 99.46%, and 99.71% for UIUC, CUReT, KTH-TIPS, UMD, and ALOT, respectively. Meanwhile, the proposed method inherits simple and efficient merits of LBP, for example, it could extract scale-robust feature for a 200×200 image within 0.24 s, which is applicable for many real-time applications.