Fusing Global Gabor Feature and Local Binary Pattern for Texture Image Recognition

Abstract

Texture is a fundamental visual feature in human vision, and texture image recognition is an important part of artificial intelligence. However, traditional texture recognition methods can’t effectively utilize global and local texture features simultaneously, which restricts the recognition performance of traditional methods. Therefore, this work combines global Gabor feature and local binary pattern to further improve the performance of texture image recognition. Firstly, a multi-scale image pyramid space is generated to reflect the scale variation of texture image. Secondly, Gabor filtering is used in the image pyramid space to extract the global Gabor feature, which is used as the global texture feature. Thirdly, the completed local binary count algorithm with multiple radii is applied to original image to extract the local binary pattern, which is used as the local texture feature. Finally, the extracted global and local texture features are fused to recognize the texture image by the nearest subspace classifier (NSC). The experimental results show that the proposed method can achieve state-of-the-art recognition accuracy with high efficiency. In addition, the proposed method is robust to scale variation and the number of training samples in texture recognition task.