多尺度细节增强与自适应γ变换的图像增强  

Image Enhancement Based on Multiscale Detail Enhancement and AdaptiveγTransformation

作　　者：孙小迎[1] 邹群艳 SUN Xiaoying;ZOU Qunyan(Development Planning Division,Nanchang Institute of Science and Technology,Nanchang 330108,China;School of Information and Artificial Intelligence,Nanchang Institute of Science and Technology,Nanchang 330108,China)

机构地区：[1]南昌工学院发展规划处,江西南昌330108 [2]南昌工学院信息与人工智能学院,江西南昌330108

出　　处：《无线电工程》2023年第6期1262-1268,共7页Radio Engineering

基　　金：江西省教育厅科学技术研究项目(GJJ212520,GJJ191095,GJJ191101,GJJ191099,GJJ202509);江西省高校人文社会科学研究项目(JY19131)。

摘　　要：针对现有图像增强方法未能适宜地提升图像的亮度、对比度,以及保持图像自然效果的问题,提出了多尺度细节增强与自适应γ变换的图像增强方法。该方法根据多尺度的纹理结构和边缘细节特征,用引导滤波对图像进行多尺度的Retinex分解,分解为多尺度的细节层和最后的基础层;对基础层做自适应的γ拉伸,实现图像亮度和对比度的有效增强,对细节层进行多尺度的拉普拉斯增强;将增强的基础层与增强的细节层进行多尺度的Retinex反变换,实现原图像的增强。图像增强实验结果显示,相对于当前的部分最新方法,所提方法的图像增强性能更好,图像增强后的信息熵和平均梯度分别比现有的方法提升大约1.2和2.8。The existing image enhancement methods fail to properly improve the brightness and contrast of the image and maintain the natural effect of the image,an image enhancement method based on multiscale detail enhancement and adaptiveγtransformation is thus proposed.This method performs multi-scale Retinex decomposition on the image based on multi-scale texture structure and edge detail features by using multi-scale guided filtering,achieves multi-scale detail layers and the last basic layer;the basic layer is enhanced by adaptiveγstretching,so as to improve the brightness and contrast of the image effectively,and the multi-scale detail layers are enhanced respectively by the improved Laplace operator;multi-scale Retinex inverse transformation with the enhanced basic layer and the enhanced layers achieves the enhanced image.The multi-scale Retinex inverse transformation is carried out on the enhanced basic layer and the enhanced layers to enhance the original image.The experimental results of image enhancement show that compared with some latest methods,the proposed method has better enhancement performance,the information entropy and average gradient of the enhanced image are about 1.2 and 2.8 higher than those of the existing methods,respectively.

关 键 词：图像增强 引导滤波 多尺度的细节层 自适应γ拉伸 改进的拉普拉斯算子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]