超声图像的LBP纹理特征提取

LBP Texture Features Extraction in Ultrasound Images

采用改进的均匀化LBP(局部二值模式)算子,将散点及小斑点的均匀化LBP码设置为其邻域均值,并通过各向异性扩散滤波抑制超声图像噪声,消除超声图像中斑点噪声和散点对纹理提取的影响,以实现对超声图像的LBP纹理特征的有效提取.实验结果表明,本文算法与传统均匀化LBP的算法相比,获得的纹理特征图像LBP码的占有比例有所提高,消除了LBP纹理图像中散点的影响,能较好地描述超声图像的纹理信息.

An improved uniform LBP（local binary pattern） operator is introduced in the paper to extract the texture features of ultrasonic images effectively. This new operator resets the uniform LBP code of scatters and small speckles to its mean of the neighborhood, then it can effectively suppress the noise and eliminate the negative effect of texture extraction by the speckle noise and scatters in ultrasound images with anisotropic diffusion. The results of experiments show that this method can get higher proportions of LBP uniform pattern and better texture information of ultrasound images than the traditional extraction methods with uniform LBP operator, and decrease the bad influence of scatters on LBP texture images.