基于增强方向局部二值模式的纹理分类

针对方向局部二值模式特征维数高且易受噪声影响的问题,提出了增强方法。首先,在方向局部二值模式的基础上,通过将局部邻域划分为多个8-正交邻域的策略,来降低方向局部二值模式的特征维数；其次,在方向局部模式中引入局部三值模式的思想,以减少噪声的影响。采用CURet、UIUC及Outex三个纹理图像库进行试验,结果证明了本文方法以较低的特征维数取得了更好的分类性能。

Texture Classification Based on Improved Direction LBP

In order to reduce the feature dimensionality and the influence of noise, an improved direction local binary pattern (DLBP) is introduced in the paper. Firstly, the neighborhood is divided into different 8-orthogonal-neighbor for dimensionality reduction of the original DLBP histogram. Then, the idea of local ternary pattern (LTP) is introduced to prevent the noise disturbance. Experimental results, obtained from three widely used texture image databases, CURet, UIUC and Outex, demonstrate that the proposed method can greatly improve the performance of the traditional DLBP on texture classification.