Fast image matching using multi-level texture descriptor

At present, image and video descriptors have been widely used in many computer vision applications. In this paper, a new hierarchical multiscale texture-based image descriptor for efficient image matching is introduced. The proposed descriptor utilizes mean values at multiscale levels of an image region to convert the image region to binary bitmaps and then applies binary operations to effectively reduce the computational time and improve noise reduction to achieve stable and fast image matching. Experimental results show high performance and robustness of our proposed method over existing descriptors on image matching under variant illumination conditions and noise.