基于空间相关性的图像分割算法研究

提出一种充分利用图像的空间相关性来达到高效快速地进行图像分割的新方法。利用均值漂移算法对图像进行分割形成过度分割的区域,并使这些区域保持理想的边缘和空间相关部分,用图结构表示的区域相邻图来代替分割的区域。和K-均值算法的思想一样,迭代循环置信传播算法以其具有收敛速度快的特点被用于最小化开销函数、整合过度分割的区域和获得最终的分割结果。基于分割区域而不是图像像素的图像聚类分割方法可降低噪声敏感性,同时提高图像分割质量。与FCM和MRF算法相比较,该算法在复杂场景图像中显示了更好的分割性能。

Image segmentation algorithm based on spatial correlation

This paper presented a full use of spatial image correlation to achieve efficient fast image segmentation method. First of all, it used mean shift image segmentation algorithm to formate an excessive segmentation, so that it made these areas to maintain the desired edge and spatial correlation part. Then, it used the graph structure of the region adjacency graph instead of segmentation. Like K-means algorithm, iterative belief propagation algorithm had the advantages of fast convergence was used to minimize the cost function, integrate over segmentation and obtain the final segmentation result. Based on the segmentation of the region rather than the image pixel, image clustering segmentation method could reduce the noise sensitivity, while improving the quality of image segmentation. Comparing with FCM and MRF algorithm, the new algorithm in entropy evaluation standard especially complex scene images shows a better performance.