南京邮电大学学报（自然科学版）2025，Vol.45Issue(2)：56-67,12.DOI:10.14132/j.cnki.1673-5439.2025.02.007
基于曲率先验改进的自适应全变分图像分割方法
Improved adaptive total variation image segmentation based on curvature prior
摘要
Abstract
Since the adaptive total variation-based image segmentation model is prone to be more sensi-tive to noise intensity and blur levels and outputs unsatisfied segmentation results,this paper proposes an improved adaptive total variation method based on the curvature prior.By improving the adaptive matrix,the influences of noise and blur in the input image on the segmentation results are reduced,which leads to relatively clear and smooth approximate images.Furthermore,the introduction of the curvature prior re-duces blocky artifacts in smooth regions.This can enhance the contrast of smooth approximate images,better preserve edges and improve the accuracy of image segmentation.A fast algorithm based on the al-ternating direction method of multipliers(ADMM)is designed to numerically solve the proposed model,and the convergence of the algorithm is verified theoretically.Experimental results on both binary and multi-phase images demonstrate the feasibility and effectiveness of the proposed method.关键词
图像分割/曲率正则化/自适应TV/交替方向乘子法Key words
image segmentation/curvature regularization/adaptive total variation/alternating direction method of multipliers(ADMM)分类
电子信息工程引用本文复制引用
易林林,张哲,闵莉花..基于曲率先验改进的自适应全变分图像分割方法[J].南京邮电大学学报（自然科学版）,2025,45(2):56-67,12.基金项目
国家自然科学基金(12271262)和南京邮电大学自然科学基金(NY221097)资助项目 （12271262）
