Abstract: Aiming at the problem of traditional methods  relying too much on local features and neglecting global features in texture classification, we proposed a feature extraction method based on the combination of local and non-local patterns.  The method integrated two algorithms: local phase quantization and non-local binary patterns. Firstly, two algorithms were used to  extract feature  from the preprocessed image separately. Secondly, the feature histograms of the two methods were weighted and fused. Finally, texture classification was performed by using the chi-square distance and the nearest neighbor classifier. In order to validate the effectiveness of the proposed method, a dataset of Manchu Eight Banners flag images was constructed, and the algorithm was applied to the classification task of the dataset. Experimental results show that, compared to single algorithm, the new algorithm has higher classification accuracy and robustness on multiple datasets.

Key words: local phase quantization, non-local binary pattern, texture classification, Manchu flag image