The Reserach Of Texture Analysis Technology Based On Adaptive Local Binary Pattern And Reaction Diffusion Filtering

Texture in our real word is various, complex and irregular. In practical application, these textures are usually affected by various factors, such as different illuminant, noise, scaling, rotation, translation and so on. Most of the existing texture feature extraction algorithms have a good result on the controlled condition (no noise or scaling elt.). While in image collection, the condition can’t be controlled easily, this may lead to the algorithms’performance drop rapidly. If we conduct some pre-treatment operations, including denoising, before feature extraction, important information, such as textures or edges, is often lost or blurred in the process of traditional smoothing algorithm. So, we mainly concern two difficulties in texture image recognition:texture image denoising and texture feature extraction in this paper.Firstly, we define a new diffusion control operator which takes the gradient as parameters combing with the wave atoms and curvelets transform, and it is used to control the velocity of the diffusion. Next, we design the double-regularizing terms which can implement the reconstruction for the lost or blurred information in diffusion process, and then propose a proposed a new texture denoising model based on reaction-diffusion equation with double-regularizing terms to balnce noise removing and texture and edge preserving. Secondly, we give a provation on the model’s convergence and compute complexity in mathematical conduction. Besides, the model’s_performace is evaluated through somg experiments on real textures images, the results and comparisons show the good performance of the proposed model for texture and edge preserving in the process of denoising especially as the noise increase.Secondly, we update the local binary pattern and define a new adaptive local binary pattern, abbreviate ALBP, and then propose a novel texture classification algorithm based on the ALBP operator to enhance the robustness under different illuminant, noise, scaling, rotation, and translation. In ALBP, we classify different texture pattern by uniformity which is used to descrip its spatial structure, and then re-label the feature combining its spatial structure and global relation factor. We obtain the difference sign matrix descriping gray-level and the difference magnitude matrix characterizing gray-intensity through differential operation, and then calculate their features respectively based on ALBP, so a local region is represented not only by its local difference sign but also through its difference magnitude. Finally, the two eigenvalues are concatenated into an enhancement feature vectors, and used for classificationFinally, we do the classification on Brodatz and Outex datasets combines the Nearest Neighborhood classifier under different illuminant, noise, scaling, rotation, and translation. Experiments show that the proposed method has better robustness than other LBP based texture classification algorithms especially under different rotation, illuminant and noise. At the same time, we also have a satisfactory reault on BeiMu images recognization.