Noise-resistant and rotation-invariant texture description and representation using local Gabor wavelets binary patterns

Abstract

This paper presents a rotation-invariant texture descriptor, which is robust to noise. In the proposed method, a given gray-scale texture image is first filtered by a set of Gabor wavelets filters. The filters are designed such that their half-peak magnitude support in the frequency spectrum touch each other with no overlap to reduce redundant information. After that a number of local binary patterns called “Local Gabor Wavelets Binary Patterns” (LGWBPs) are computed based on the obtained Gabor wavelets filters responses via global measures. The histogram of the computed LGWBPs is then used as a texture feature vector. Extensive experiments were conducted on the well-known Outex, and CUReT databases in the presence of different levels of Gaussion noise. Experimental results indicate that the proposed method can be utilized as a suitable noise-robust and rotation-invariant texture descriptor for texture classification.