一种基于Contourlet变换和视觉感知特性的图像分割方法

An Image Segmentation Algorithm Based on Contourlet Transform and Vision Perception Information

纹理图像的分割通常是通过分离块或移动块进行的,即以块的特征作为块或块中心像素的特征来进行,纹理边界和图像边界像素的误分率高.提出一种新的基于视觉感知特性邻域结构的纹理图像分割方法.该方法分为两步:首先用移动块产生中心像素的特征属性值;然后每个像素建立9个特殊邻域结构窗口,通过邻域结构中属性特征的一致性来确定像素的特征属性值.用此方法对30种纹理图像进行了分割实验,取得了满意的结果.

Texture image segmentation is implemented by separating the image into distinct blocks or sliding blocks. The texture values were assigned to a pixel by using a window centered about that pixel. There are some considerable problems that can occur at the boundary of the image, the border between different textures in an image. This paper analyses those problems and propose a solutions that determined the pixel＇s feature value by using the vision perception information of pixel＇s neighborhood structure. This method includes two steps. Firstly, the pixels＇ feature value is assigned to a texture value of the window centered about it. And then we use the value of the highest homogeneity of nine special neighborhoods by the local structure as the pixel＇s. The analysis and experimental results on 30＇s synthesis texture image demonstrate the effectiveness of this algorithm.