基于自适应分割和多尺度Retinex的图像增强算法  

Image enhancement algorithm based on adaptive segmentation and multi-scale Retinex

作　　者：李雅梅[1] 谢秉旺 LI Yamei;XIE Bingwang(School of Electrical and Control Engineering,Liaoning Technical University,Huludao 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：《传感器与微系统》2023年第10期135-138,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(51974151,71771111);辽宁省教育厅基金资助项目(LJ2019JL013)。

摘　　要：智能行车预警是智慧交通系统(ITS)的主要组成之一,因车载设备采集图像受环境影响,会出现强光影过渡和局部高亮的现象。针对Retinex算法在该场景下存在光晕伪影与纹理弱化的问题,提出了一种基于自适应分割和多尺度Retinex(MSR)的图像增强算法。首先,使用中值滤波器滤除基本噪声,同时保留边缘信息,再通过以景深为主导的改进分水岭分割将图像分割为亮、暗2个独立区域;其次,对亮、暗区分别进行不同尺度的Retinex增强,根据亮、暗区域的边缘特性,计算融合权重并进行亮、暗区的融合重构,得到最终的增强图像。实验结果表明:所提方法在强光影过渡和低照度逆光图像的增强实验中具有良好的增强效果,抑制过渡区光晕的同时增强了高亮区的细节。Intelligent driving early warning is one of the main components of intelligent transportation system(ITS).Due to the influence of the environment on the image collected by vehicle equipment,strong light transition and local highlighting will occur.An image enhancement algorithm based on adaptive segmentation and multi-scale Retinex(MSR)is proposed aiming at the problem of halo artifact and texture weakening of Retinex algorithm in this scene.Firstly,the median filter is used to remove the basic noise and retain the edge information at the same time.Then,the image is divided into light and dark regions by improved watershed segmentation based on the depth of field.Secondly,Retinex enhancement of different scales is performed on the bright and dark areas respectively.According to the edge characteristics of the bright and dark areas,the fusion weights are calculated and the fusion and reconstruction of the bright and dark areas is carried out to obtain the final enhanced image.Experimental results show that the proposed method has good enhancement effect in strong light transition and low illumination backlight image enhancement experiments,and can suppress the halo in the transition region and enhance the details in the highlight region.

关 键 词：图像处理 RETINEX算法 分水岭分割 图像融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]