基于均值偏移的灰度图像分割方法

Gray Image Segmentation Algorithm Based on Mean Shift

针对灰度图像分割往往存在的过分割或欠分割问题,提出了一种基于均值偏移的图像分割方法.该方法通过联合像素的空间位置和灰度特征,建立图像的特征矢量,构造基于像素的位置和灰度的改进核函数直方图.采用均值偏移算法搜索图像的特征模式,并以此对图像进行平滑滤波和分割.对以地物为背景的图像分割结果表明,该方法既能抑制具有纹理的大片背景,又能提取出面积较小且轮廓清晰的物体,分割的物体完整且符合人眼视觉,较有效地避免了图像过分割和欠分割的问题.

The new image segmentaiotion algorithm based on mean shift is proposed to overcome such problems as over-segmentation or lack-segmentation, and to improve the performance of gray image segmentation. It constructs a novel kernel function histogram by combing the position and the gray value of the pixel, and then makes use of mean shift with this new kernel function to automatically search the models of the features, and to segment the gray image. Experiments based on the ground as the main background image segmentation are carried out by the Canny,Sobel and this mean shift,and the results show that the proposed algorithm not only suppress the big object and small object effectively,and it could effectively avoids image over-segmentation or lack-segmentation.