Image local feature extraction in complex illumination scenes has always been a research hotspot of image processing. For the simple quantization method of WLD operator and the lack of directional feature extraction, this paper proposes a new image local feature descriptor called Anisotropy Weber Local binary Pattern (AWLBP). The differential excitation component in the WLD operator is replaced by an anisotropic LOG operator that introduces the scale parameter and the angle parameter. The gradient directios component is replaced by the LBP operator. Combine the two to generate a two-dimensional AWLBP histogram, which is then converted into a one-dimensional histogram, and finally classified using the KNN classifier. The algorithm validates its validity and accuracy in a large number of experiments in the CMUPIE face database and the PhoTex texture image library. The experimental results show that the proposed image feature extraction algorithm is highly effective and robust in complex illumination scenarios.