融入图像块和类信息量的无损检测图像分割

Image Segmentation of Non-Destructive Test Based on Image Patch and Cluster Information Quantity

在图像分割中,考虑邻域信息的模糊C均值算法能够有效地降低噪声对图像的干扰,但这类算法需额外引入参数,且无损检测图像的较大类间差异易导致分割失败。为此,提出基于图像块的类间差异不敏感的模糊C均值算法。利用像素所在的图像块代替像素进入聚类进程,图像块内像素的权重由像素的空间距离和灰度大小自适应确定。基于信息量的概念,给出类信息量表征形式并将其引入目标函数以改善常见模糊C均值算法对类间差异敏感的缺陷。基于新构建的目标函数得到新的隶属度和聚类中心表达式,并给出算法流程。最后,利用类间差异较大的无损检测图像对所提算法进行测试,结果表明:与其他模糊聚类算法相比,本文算法具有更高的分割准确率和更好的视觉效果。

In image segmentation,many fuzzy C-means algorithms considering neighborhood information can effectively reduce noise interference,but these algorithms need additional parameters,and the large cluster difference between nondestructive test images easily causes segmentation failures.To solve this problem,this paper presents a fuzzy C-means algorithm insensitive to cluster difference based on image patchs.First,the image patch is used to replace the pixel to enter the clustering process.The weight of the pixel in the image patch is adaptively determined by the spatial distance and gray scale of the pixels.Second,based on the concept of information quantity,the expression of cluster information quantity is given and introduced into the objective function to improve the sensitivity of common fuzzy C-means algorithms to the cluster difference.Third,the new expressions of membership degree and cluster center are obtained based on the new objective function,and the algorithm flow is given.Finally,the proposed algorithm is tested by the non-destructive test images with large cluster difference.The results show that the proposed algorithm has high segmentation accuracy and better visual effects.