边缘信息指导下的半模糊聚类图像分割方法

提出了一种利用边缘信息的半模糊均值聚类的图像分割算法，它先用边缘检测和区域生长算法对图像进行一次预分割，确定聚类的初始参数，然后在这个基础上对“边缘”部分的点采用模糊聚类、非“边缘”部分使用分明聚类，避免了模糊聚类时初始参数设定的盲目性，减少了迭代时的计算量，提高了迭代收敛速度．除灰度特征外，聚类时还利用了点到类的距离特征，较好地保持了分割图像的连续性．直接观察对比多幅图像的分割实验结果可以明显地发现，该算法较常用的Cksu方法、二维熵阈值分割方法以及FCM方法的分割结果更准确．就Lena图像而言，该算法的收敛速度也比一般的FCM快了将近一倍．

A semi-fuzzy clustered image segmentation method under edge detection

Image segmentation is an important and elementary part of image analysis and pattern recognition. Many researchers are working on the improvement of quality and reducing the complexity of segmentation. A new hybrid method for image segmentation incorporating edge info and clustering was presented. In this approach, a rough segmentation using the edge detection and region-growing was made to work out the initial parameters of clustering, reducing the blindness for parameters of clustering, decreasing the complexity dramatically. In order to keep the continuity of each sub-image, the distance between pixel and cluster is used when iterating. Compared intuitively with the results by using some common methods such as Ostu, regular FCM and two-dimensional entropy threshold algorithm, it was seen that the method is more accurate. The convergence speed of the method is almost two times as fast as that of the regulation FCM with the Lena image, the validation and accuracy of the method was proved.