Completed hybrid local binary pattern for texture classification

Abstract

The Local Binary Pattern (LBP) and its variants have been widely investigated in image processing and computer vision applications, e.g., texture classification due to their powerful ability to capture image features and computational simplicity. However, owing to the simple selection strategy of the threshold, the original LBP descriptor is sensitive to noise and illumination variations and tends to characterize different local patterns with the same binary code. Recently, the Completed Robust Local Binary Pattern (CRLBP) has been introduced to overcome these demerits, in which the Weighted Local Gray Level (WLG) is introduced to replace the traditional gray value of the center pixel, but the improvement is not significant and one additional parameter has to be tuned. To address these difficulties effectively, this paper proposes a hybrid framework of LBP, called Completed Hybrid Local Binary Pattern (CHLBP), in which a first order derivative and a second order derivative are combined to represent local patterns. In order to make CHLBP more robust and stable, more relationship information among pixels in the local region is exploited, that is, the Average Local Gray Level (ALG) is adopted to take place of the traditional gray value of the center pixel as well as the neighbor pixels. The results obtained from two representative texture databases show that the proposed method is robust to illuminant variations and viewpoint variations and can achieve impressive classification accuracy. The proposed model improves the classification results from 96.95% to 98.78% on the Outex database, and from 91.85% to 94.56% on the UIUC database as compared with the Completed Local Binary Pattern (CLBP), which is the benchmark method of LBP-based models.