基于高斯局部二值模式的纹理特征分类方法

摘要

The texture description of machine vision is important for image analysis and pattern recognition. To improve the robustness of feature describing, a texture feature based on Gaussian local binary pattern(LBP) is proposed in this paper. Firstly, the Gaussian filtering is used to construct the multi-scale images as the image pyramid after image enhancement. Secondly, the local binary pattern is improved to enhance the rotation invariance and noise immunity. The pattern is extracted by using the mean value and the principal direction. Finally, for the different scales, the feature vector of three local binary patterns is extracted and reduced by histogram for image classification. The experiment result shows the feature has good distinguishability and efficiency and it is applicable for image classification.%纹理描述在图像分析和模式分类领域具有极为重要的意义.为提高特征描述的鲁棒性,提出了一种基于高斯局部二值模式的纹理特征提取方法.首先,采用加强预处理使高斯滤波获得不同尺度的图像,从而构建多尺度的图像金字塔;其次,为提升旋转不变性和抗噪声能力,提出具有主方向特征的二值模式;最后,在不同尺度上提取3种有效的局部二值模式联合构造纹理描述,并通过直方图降维.试验结果表明,该特征具有较好的可区分性和有效性,可以有效应用到视觉图像的纹理分类中.