Texture matching using local and global descriptor

Image processing is one of the important areas of research, which provides efficient solutions to many real and industrial problems. Texture analysis is the most important field in image processing because all objects are textured in real world. In this work, we propose a new texture segmentation method based on the dynamic segmentation architecture. This architecture decomposes the image into blocks with the same size of a main window. After that, the size of the main window is reduced and the same process is applied to extract other blocks with different size. This process is repeated until the size of the main window reaches a minimum size. Neuroscience studies said that the human brain combines between Local and Global features to recognize objects. Based on these studies, the feature extraction step of the proposed segmentation method is based on the combination between two methods. The first method extracts the local feature using the Local binary patterns (LBP) method. And the second method captures the global information of the texture using Radon transform. Synthesis images and generated images from Brodatz album database have been used in the evaluation part. The experiments illustrate the efficiency and the improvement made by the proposed combination, compared to [1] and [2], to extract the local and directional information of the texture.