Antinoise Rotation Invariant Texture Classification Based on LBP Features of Dominant Curvelet Subbands
Yan Shang, Weimin Hou, Ruihong Wu, Zhi-Yong Meng · 2008
Instead of computing the LBP histogram of texture image in space directly which has some limitations to classification, a rotation invariant texture classification algorithm based on the multiresolution LBP features of dominant curvelet subbands in the combination of space and frequency domain is proposed. The texture image is transformed by curvelet first, then compute the LBP histogram of the resampled image that is reconstructed using dominant directional subbands of each scale. The rotation invariant feature vectors have the multiresolution and antinoise properties, the LBP operators of the same size can character the original texture in larger region so as to avoid the disadvantage of traditional LBP. The images are classified by support vector machines (SVM) at last. The proposed method is compared with other texture classification algorithm, the experiment results show that it can improve classification rate effectively and have stronger antinoise properties.