结合非局部均值的快速FCM算法分割MR图像研究

Research on MR Image Segmentation Based on Fast FCM Algorithm Combined with Non-local Means

针对FCM算法分割医学MR图像存在的运算速度慢、对初始值敏感以及难以处理MR图像中固有Rician噪声等缺陷,提出了一种结合非局部均值的快速FCM算法。该算法的核心是首先针对MR图像中存在的Rician噪声,利用非局部均值算法对图像进行去噪处理,消除噪声对分割结果的影响;然后根据所提出的新的自动获取聚类中心的规则得到初始聚类中心;最后将得到的聚类中心作为快速FCM算法的初始聚类中心用于去噪后的图像分割,解决了随机选择初始聚类中心造成的搜索速度慢和容易陷入局部极值的问题。实验表明,该算法能够快速有效地分割图像,并且具有较好的抗噪能力。

Aiming at the drawbacks existing in the FCM algorithm of the slow speed of operation,result vulnerable to the initial value and the difficulty to deal with the inherent Rician noise of MR image,this paper presented a fast FCM algorithm combined with non-local means.The core of the algorithm is as follow.Firstly,aiming at the Rician noise existing in the MR image,using the non-local means algorithm to deal with the noise,eliminating the impact of noise on segmentation result.Secondly,getting the initial cluster centers automatically according to the proposed rules of initial centers.Finally,the cluster centers should be as the initial cluster centers of fast FCM for the segmentation of the denoised image to solve the slow search speed and the problem that is easy to fall into local minima caused by the random selection of the initial clusters.Experimental results show that the proposed algorithm can quickly and efficiently segment the image,and is more robust to noise.