Texture Image Classification with Noise-Tolerant Local Binary Pattern

The local binary pattern (LBP) is a simple and effective texture descriptor. However, it is very sensitive to image noise. To deal with this problem, we propose an efficient texture feature named noise-tolerant complete enhanced local binary pattern (CELBP\+{NT}) to enhance the discriminant ability against the noisy texture images. Derived from the local binary pattern, CELBP\+{NT} is robust to illumination, rotation and noise. Its feature extraction process involves the following three steps. First, different patterns in LBP are reclassified to form an enhanced LBP (ELBP) based on their structures and frequencies. Then, in order to describe the local feature completely and sufficiently, the difference of modulus value and center pixel information is added to ELBP to develop a complete ELBP feature, named CELBP. Meanwhile, the adaptive threshold of CELBP is determined by the image size. Finally, CELBP\+{NT} is proposed by using the favorable characteristics of multi-scale analysis on CELBP. The features are evaluated on the popular Outex database with different intensity and different types of noise. Extensive experimental results show that CELBP\+{NT} not only demonstrates better performance to a number of state-of-the-art LBP variants under no-noise condition, but also effectively improves the performance of texture classification containing noise due to its high robustness and distinctiveness.