基于不同色彩空间融合的快速图像增强算法

针对现有Retinex算法中存在的色彩失真、噪声放大及光晕伪影现象等问题，本文提出了一种基于Retinex理论的改进算法. 该算法首先在HSV空间对亮度分量V通道进行增强处理，同时在拉伸得到的对数域反射分量至一定的动态范围时（本文是0～255），引入增强调整因子，调整不同亮度值的增强程度来避免噪声放大及色彩失真现象；然后在RGB空间，通过分析光晕产生的原因，提出一种改进的高斯滤波器来消除光晕现象，并在计算反射分量时，通过参数调整图像颜色的保真度. 最后，对上述两种不同颜色空间的处理结果进行加权平均作为算法的最终输出. 实验结果表明，针对不同光照条件下的图像，1）该算法可以明显地改善光晕伪影现象；2）无色彩失真、噪声放大等问题；3）效果和效率优于带色彩恢复的多尺度Retinex算法（Multi-scale retinex with color restoration，MSRCR）及其他对比算法.

A Fast Image Enhancement Algorithm Based on Fusion of Different Color Spaces

A fast image enhancement algorithm based on fusion of different color spaces is proposed to overcome the problems of color distortion, noise amplification, and halo artifacts. Firstly, a single scale retinex is used in channel V of the HSV color space. An enhancement adjustment factor is introduced, when we stretch the calculated reflection in the logarithmic domain into a dynamic range, i.e., 0～255. It is used to adjust the enhancement for different pixels. Thus noise amplification and color distortion can be effectively avoided. Then, the improved Gaussian filter is given in the RGB color space by analyzing the cause of the halo effect. This step can eliminate the halo artifact. And a parameter is adopted to keep the color natural of the image when the reflection is calculated. Finally, the weighted average of the outputs of the above two color spaces is taken as the final output of our algorithm. The experiment results show that for images with different lighting conditions, 1) the outputs of our algorithm is free from the halo artifacts; 2) there are no color distortion and noise amplification problems; 3) the quality and the efficiency of the algorithm are superior to the multi-scale retinex algorithm (MSRCR) and other comparison algorithms.