基于金字塔分解和扇形局部均值二值模式的鲁棒纹理分类方法

针对传统局部二值模式(LBP)的特征鉴别力有限和噪声敏感性问题，该文提出一种基于金字塔分解和扇形局部均值二值模式的纹理特征提取方法。首先，将原始图像进行金字塔分解，得到对应于不同分解级别的低频和高频(差分)图像。为提取兼具鉴别力和稳健性的特征，进一步采用阈值化处理技术将高频图像转化为正、负高频图。然后，基于局部均值操作提出一种扇形局部均值二值模式(SLMBP)，用于计算各级分解图像的纹理特征码。最后，对纹理特征码进行跨频带的联合编码和跨级别的直方图加权，从而获得最终的纹理特征。在公开的3个纹理数据库(Outex, Brodatz和UIUC)上进行分类实验，结果表明该文所提方法能够有效地提高纹理图像在无噪声环境和含高斯噪声环境下的分类精度。

Robust Texture Classification Method Based on Pyramid Decomposition and Sectored Local Mean Binary Pattern

The traditional Local Binary Pattern (LBP) has limited feature discrimination and is sensitive to the noise. In order to alleviate these problems, this paper proposes a method to extract texture features based on pyramid decomposition and sectored local mean binary pattern. First, the pyramid decomposition is performed on the original image to obtain low-frequency and high-frequency (difference) images with different decomposition levels. To extract robust yet discriminative features, thresholding technique is further used to transform the high-frequency images into positive and negative high-frequency images. Then, based on local averaging operations, Sectored Local Mean Binary Pattern (SLMBP) is proposed and used to compute texture feature codes at different decomposition levels. Finally, the texture features are obtained by joint coding across frequency bands and by histogram weighting across decomposition levels. Experiments on three publicly available texture databases (Outex, Brodatz and UIUC) demonstrate that the proposed method can effectively improve the classification accuracy of texture images both in noise-free conditions and in the presence of different levels of Gaussian noise.