一种改进的二维直方图均值漂移分割算法

Image segmentation of mean shift based on improved 2D histogram

改进了二维直方图的构造方法,利用空间邻域信息使改进的二维直方图具有更丰富的噪声判断信息,并根据此信息将图像分为噪声子图和非噪声子图。采用均值漂移算法对图像进行聚类分割,并对均值漂移的高斯核函数进行了改造,使算法对噪声有更好的平滑作用,对非噪声区域有更准确的分割效果。实验结果表明,改进的算法对噪声污染的图像有更好的抗噪能力,分割也更加准确。

This paper improved the establishment of 2D histogram to contain more detailed information of noise estimation, according to which the image was divided into noise sub-image and non-noise sub-image. Then mean shift was applied to cluster and segment. The modified Gaussian kernel function could smooth noise better and cluster more accurately in non-noise subimage. The experimental results show that the improved algorithm has better performance of anti-noise and more accurate precision.