NSCT变换与空间约束模糊核聚类红外图像分割算法

红外图像成像模糊、易受噪声污染,分辨率低,采用标准的FCM分割算法会出现失效和误分割。通过对以往各种方法的研究,根据红外图像的特点及FCM算法的不足,提出采用在NSCT变换域进行去噪预处理与改进的FCM算法相结合的分割算法。首先对红外图像进行NSCT变换,在变换域,采用自适应阈值法去除各细节子带中的噪声,其次在FCM算法中引入核映射将数据映射到非线性空间中进行聚类划分,最后采用邻域信息修正当前像素的隶属度值,得到更准确的聚类结果。实验结果证明该算法较FCM、KFCM、SFCM聚类分割算法有更好的分割精度。

NSCT transform and improved Kernel fuzzy c-means clustering with spatial constraints for infrared image segmentation

The membership is used in the standard Fuzzy c-means algorithm for image soft segmentation, especially for the image with blur edges. FCM is not suitable for the low contrast and noisy infrared image. In order to achieve the correct segmentation, a new algorithm is proposed. First step is removing noise by the NSCT. The next step is clustering by the modified FCM. The data is mapped to non-linear space for easy clustering by kernel function. The neighborhood information is used for correcting the membership of every pixel. Experiments result has been show that the algorithm has higher accuracy and better segmentation quality than others.