模糊域内基于Retinex的雾霾图像增强算法  

Foggy and Hazy Image Enhancement Algorithm Based on Retinex in Fuzzy Field

作　　者：贾伟[1] 刘彦宾[1] 刘炜[1,2] 葛耿育[1] 苏文丽[1] 樊里略[1] 

机构地区：[1]遵义师范学院计算机与信息科学学院,遵义563002 [2]贵州大学计算机科学与技术学院,贵阳563003

出　　处：《计算机科学》2015年第B11期183-188,共6页Computer Science

基　　金：贵州省科技厅;遵义市科技局;遵义师范学院联合基金项目(黔科合J字LKZS[2014]23号;黔科合J字LKZS[2014]08号);遵义师范学院教学科研项目(13-08;13-06)资助

摘　　要：现有Retinex图像增强算法在使用过程中往往会产生轻微的光晕现象,在图像清晰度、细节、保真性、适用范围等方面存在诸多不足,在模糊域内,提出的基于Retinex的雾霾图像增强算法可有效克服此类不足。首先利用自适应多阈值算法对图像进行分块,并确定分块区域的最佳渡越点;然后采用线性隶属度函数将图像像素值变换为模糊域,通过渡越点计算提出的模糊双曲正切函数的关联参数,对图像的各个分块区域应用Retinex算法进行非线性图像增强,同时对增强结果进行模糊双曲正切调整;最后采用线性加权和线性逆变换的方法恢复原图像。通过对大量图像的对比实验发现:传统处理方法产生的光晕现象得到了抑制,图像清晰度、细节、保真性以及对比度等处理效果改善明显,算法适用范围更广。The traditional Retinex algorithm always products ＂halo＂ effect in fog and haze image enhancement, and has several shortcomings, such as the poor image exposure, image sharpness, image details, image fidelity and so on. In order to overcome these shortcomings of the traditional algorithm, we proposed a foggy and hazy image enhancement algorithm based on Retinex in the fuzzy field. Firstly, the original image was classified into several blocks using the proposed adaptive multi-threshold algorithm, and then the optimal crossover points of the image blocking areas were computed. Secondly,we used a novel linear membership function to map the image pixel value into fuzzy domain, then computed the correlation parameter of the fuzzy hyperbolic tangent function by crossover points, used the Retinex algorithm to perform no-linear image enhancement, and used the fuzzy hyperbolic tangent functions to adjust the enhancement result. Finally, the method of superposition of the linear was used to map the enhancement result into original image domain. The experimental result shows that the ＂halo＂ effect is suppressed, the proposed method plays a better role in the image sharpness, image details and image fidelity, and it has more widely applicability.

关 键 词：模糊域 雾霾图像增强 RETINEX算法 模糊双曲正切 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]