期刊导航

论文摘要

Retinex模型下基于融合策略的雾霾图像增强

Haze Image Enhancement Based on Fusion Strategy in Retinex Model

作者:李昌利(河海大学 计算机与信息学院, 江苏 南京 211100);周晓晓(河海大学 计算机与信息学院, 江苏 南京 211100);张振(河海大学 计算机与信息学院, 江苏 南京 211100);樊棠怀(南昌工程学院 信息工程学院, 江西 南昌 330099)

Author:LI Changli(College of Computer and Info., Hohai Univ., Nanjing 211100, China);ZHOU Xiaoxiao(College of Computer and Info., Hohai Univ., Nanjing 211100, China);ZHANG Zhen(College of Computer and Info., Hohai Univ., Nanjing 211100, China);FAN Tanghuai(School of Info. Eng., Nanchang Inst. of Technol., Nanchang 330099, China)

收稿日期:2017-06-12          年卷(期)页码:2018,50(5):202-208

期刊名称:工程科学与技术

Journal Name:Advanced Engineering Sciences

关键字:雾霾图像;Retinex模型;色彩失真;图像增强

Key words:haze image;Retinex model;color distortion;image enhancement

基金项目:国家自然科学基金资助项目(61871174;61563036)

中文摘要

针对现有Retinex算法处理后的图像达不到色彩与细节同时增强的效果,且细节增强的同时易产生光晕,出现噪声放大、色彩失真等情况,提出了一种Retinex模型下基于融合策略的雾霾图像增强算法。该算法首先在HSV空间实现色彩增强,根据亮度分量,获取无色彩失真的反射分量,再通过修正照度分量,得到改进的Retinex模型;然后在RGB空间,采用快速双边滤波器来保留图像细节,在获得反射分量的基础上,引入原图部分和色彩恢复因子,实现细节增强;最后,在RGB色彩空间对处理后的图像加权融合,得到增强后的去雾图像。通过本文算法和现有算法对雾天图像进行去雾处理,得到不同的去雾结果。本文算法得到的去雾图像增加了细节信息,没有出现明显的颜色失真和光晕现象,处理后图像的信息熵提高。本文算法具有更短的运算时间,具有良好的可操作性。根据实验结果以及评价标准,本文算法能够在减少色彩失真的基础上,达到细节增强的良好效果。

英文摘要

In order to resolve the problem that the existing Retinex algorithm for the image processed could not achieve the enhancement of color and detail at the same time, and the enhancement of detail easily produced halo, noise amplification, color distortion, etc., a haze image enhancement algorithm based on the fusion strategy of Retinex model was proposed. Firstly, the color enhancement was achieved in HSV space, and the reflection component without color distortion was obtained according to the luminance component. Secondly, in the RGB space, the fast bilateral filter was used to preserve the detail of the image. On the basis of obtaining the reflection component, the original part and the color recovery factor were introduced to realize the detail enhancement. Finally, the processed image was weighted and fused in the RGB color space and then the dehazed image was obtained. The haze image which was dehazed by the proposed algorithm and the existing algorithm could obtain different dehazing results. The dehazed image obtained by the proposed algorithm increases the detail information, and there is no obvious color distortion and halo phenomenon, and the information entropy of the processed image is improved. This algorithm has shorter computation time and good operability. According to the experimental results and evaluation criteria, the proposed algorithm could achieve a good effect of detail enhancement on the basis of reducing color distortion.

关闭

Copyright © 2020四川大学期刊社 版权所有.

地址:成都市一环路南一段24号

邮编:610065