融合灰度相对变换和非局部均值的图像分割  

Image Segmentation Based on Gray Relative Transformation and Non-local Mean

作　　者：兰文宝 车畅 LAN Wen-bao;CHE Chang(School of Technology,Harbin University,Harbin 150001,China)

机构地区：[1]哈尔滨学院工学院,黑龙江哈尔滨150001

出　　处：《控制工程》2020年第10期1726-1735,共10页Control Engineering of China

基　　金：2019年度黑龙江省高等教育教学改革研究项目(SJGY20190397)。

摘　　要：针对非均匀光照图像分割精度不高等问题,提出了一种融合灰度相对变换和非局部均值的模糊熵分割算法。首先利用非局部均值滤波对图像进行去噪,将像素与局部极值的相对值代替像素本身灰度值,根据灰度相对变换求取局部隶属度,引入Sobel算子修正隶属度,降低灰度跳变对分割的影响;借助非局部空间信息整合水平-垂直两个方向建立隶属度矩阵,继而进行直觉模糊化形成直觉模糊熵;最后利用最小熵值法分割图像。仿真实验表明,本文提出的分割算法取得了较好的分割效果,并在错分率、过分割和欠分割等评价指标上至少提高13.59%、3.18%和2.48%。To solve the problem of low precision of non-uniform illumination image segmentation, a fuzzy entropy segmentation algorithm is proposed, which combines gray-scale relative transformation and non-local mean. First, the image is denoised by non-local mean filtering, the relative value of pixel and local extreme value is substituted for the gray value of pixel itself, the local membership degree is obtained according to the relative transformation of gray scale, Sobel operator is introduced to modify membership and reduce the influence of gray jump on segmentation. Integrating horizontal and vertical directions with non-local spatial information, the membership matrix is established, and then intuitionistic fuzzy entropy is formed by intuitionistic fuzzy. Finally, the image is segmented by minimum entropy method. Simulation experiments show that the proposed segmentation algorithm has achieved good segmentation results and improved at least 13.59%, 3.18%, and 2.48% on evaluation indexes such as misclassification rate, over-segmentation, and under-segmentation, respectively.

关 键 词：非均匀光照 图像分割 灰度相对变换 非均值滤波 边缘检测 直觉模糊熵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]