基于梯度域L_(2)正则化重建模型的边缘感知图像处理方法  

Gradient Domain L_(2 )Regularized Reconstruction Model for Edge-aware Image Processing

作　　者：孙宁康 杨洋[2] 郑好 SUN Ningkang;YANG Yang;ZHENG Hao(School of Physics and Electronic Engineering,Jiangsu University;School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]江苏大学物理与电子工程学院 [2]江苏大学计算机科学与通信工程学院,江苏镇江212013

出　　处：《软件导刊》2023年第12期192-199,共8页Software Guide

基　　金：国家自然科学基金项目(61402205)。

摘　　要：边缘感知图像处理是计算机图形学领域的一个重要课题,在图像细节增强和色调映射中应用颇多,然而现有处理方法常常受到光晕及梯度反转伪影的影响。为此,提出一种新颖的映射函数,其可在保持边缘的同时对图像细节进行灵活处理,从而适用于多种应用。同时提出一种基于梯度域L2正则化的重构模型,用于边缘感知图像处理。该模型从映射的梯度中重建处理后的图像,可以显著减轻光晕以及梯度反转伪影的影响;且该模型可基于傅立叶变换快速有效求解,在Intel i7-6700 CPU上处理一幅100万像素图像耗时0.46 s。定性定量实验结果表明,该图像处理方法在图像增强和色调映射中取得了较好结果,细节增强指标SSEQ为11.39,ILNIQE为27.52;色调映射指标SSEQ为14.88,ILNIQE为22.78。Edge perception image processing is an important topic in the field of computer graphics,with many applications in image detail enhancement and tone mapping.However,existing processing methods are often affected by halo and gradient reversal artifacts.To this end,a novel mapping function is proposed,which can flexibly process image details while maintaining edges,making it suitable for various applica⁃tions.Simultaneously,a reconstruction model based on gradient domain regularization is proposed for edge perception image processing.This model can significantly reduce the impact of halo and gradient reversal artifacts by reconstructing the processed image from the mapped gradi⁃ent;And this model can be quickly and effectively solved based on Fourier transform,with a processing time of 0.46 seconds for a 1 megapixel image on the Intel i7-6700 CPU.The qualitative and quantitative experimental results show that the image processing method has achieved good results in image enhancement and tone mapping,with detail enhancement indicators SSEQ of 11.39 and ILNIQE of 27.52;The tone map⁃ping indicators SSEQ are 14.88 and ILNIQE are 22.78.

关 键 词：边缘感知图像处理 映射函数 L_(2)正则化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]