基于双域融合与亮度补偿的室内图像增强算法  

Indoor Image Enhancement Algorithm Based on Dual-Domain Fusion and Brightness Compensation

作　　者：徐耀松[1] 刘翰林 王雨虹[1] XU Yaosong;LIU Hanlin;WANG Yuhong(Faculty of Electrical and Control Engineering,Liaoning Technical University,Huludao Liaoning 125105,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,辽宁葫芦岛125105

出　　处：《传感技术学报》2023年第1期45-52,共8页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金项目(51974151,71771111);辽宁省自然基金指导计划项目(LJ2019QL015)。

摘　　要：针对室内环境下的图像易存在失真、整体偏暗、点光源区域明暗交接处存在光晕效应、边缘细节丢失等问题,提出一种基于双域融合与点光源亮度补偿模型相结合的室内图像增强算法。该算法首先采用Canny算子边缘检测方法对图像进行分割,区分出明亮区域和非明亮区域,然后使用亮通道先验算法和暗通道先验算法粗算明亮区域和非明亮区域的透射率后进行正则化求解,用含有归一化参数的点光源亮度补偿模型进行最优步长迭代优化透射率,使图像透射率具有自适应调节能力。黄玉川煤矿选煤厂中监控图像具有典型的失真、偏暗、点光源影响大等特点,以此应用背景为例,对所提出的算法进行实验验证,结果表明处理后图片在峰值信噪比(PSNR)平均提高约20%、平均梯度(MG)平均提高约25%、信息熵(IE)平均提高约30%、结构相似度(SSIM)平均提高约25%,在减少图像明暗交接处光晕、细节保持、鲁棒性等方面均有显著提升。Aiming at the problems of image distortion in indoor environment,overall darkening,halo effect at the intersection of light and dark in point light source area,loss of edge details,etc.,an indoor image enhancement algorithm based on the combination of dual-domain fusion and point light source brightness compensation model is proposed.Firstly,the Canny operator edge detection method is proposed to segment the image,and distinguish bright areas and non-bright areas.The bright channel prior algorithm and the dark channel prior algorithm are adopted to roughly calculate the transmittance of bright area and non-bright area,then the solution is regularized,and the point light brightness compensation model with unitary parameters is adopted to iteratively optimize the transmittance with optimal step,which makes the image transmittance have self-adaptive adjustment ability.The monitoring images in the coal preparation plant of Huangyuchuan Coal Mine have the characteristics of typical distortion,partial darkness,and great influence of point light sources.Taking this application background as an example,the algorithm proposed is experimentally verified.The results show that the peak signal-to-noise ratio(PSNR)of the processed images is improved by about 20%on average,the mean gradient(MG)is improved by about 25%,the information entropy(IE)is improved by about 30%,and the structural similarity(SSIM)by about 25%.Significant improvements have been made in reducing halo at the junction of image light and dark,detail preservation,and robustness.

关 键 词：室内图像增强 双域融合 Canny算子边缘检测 光晕效应 点光源亮度补偿模型 正则化 大气散射模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]