A novel texture image pyramid based vote strategy in local binary pattern for texture classification

In texture classification field, Local Binary Pattern (LBP) is one of the most attractive method because of its robust texture feature extracting capability and low computational cost. However, there are two main shortages need to be solved. Firstly, original LBP extracts texture features in a fixed circle temple and cannot extract multi-scale texture features. Secondly, the recent researches only combine the feature histograms to utilize the complementary texture features extracted by different LBP-based variants. In this paper, aiming to solve these two main shortages, we propose a novel texture image pyramid based vote (TIPbV) strategy in local binary pattern. There are totally three steps in the TIPbV strategy: the preprocessing step, the vote step and the complementary texture feature extraction step. In the first preprocessing step, we build a multi-scale Gaussian texture image pyramid or multi-scale average texture image pyramid from original texture images in different texture databases. In the second vote step, the unanimous vote strategy or the majority vote strategy is utilized to effectively utilize multi-scale texture features extracted by original LBP. In the third complementary texture feature extraction step, we will utilize LBP-based methods to extract the complementary texture features from original texture images. The final training image which has most similar multi-scale and complementary texture features with the testing image will be selected to determine the classification result.