改进的多尺度Retinex算法及其应用

在经典的多尺度Retinex算法中对Retinex输出采用一个常数增益，使图像在平滑区域和高对比度边缘出现过增强，导致噪声放大和边缘晕环。针对该问题，提出改进MSR算法，对Retinex输出采用自适应空间变化增益，平滑区域和高对比度边缘增益小，细节区域增益大，并且小尺度Retinex输出不同区域增益差大，而大尺度Retinex输出不同区域增益差小，从而使图像细节更清晰，同时场景轮廓和颜色呈现更自然。将该算法用于受到严重退化的雾天图像，能取得较好的图像去雾效果。

Improved Multi-scale Retinex Algorithm and Its Application

In the standard multi-scale Retinex algorithm,a constant gain is applied to a Retinex output,which leads to overenhancement in smooth and edge regions,in which noise amplification and ringing artifacts take place,respectively.An improved Multi-Scale Retinex（MSR） algorithm is proposed by applying the adaptive space varying gain,which means larger gain is applied to pixels in smooth and edge regions while smaller gain is applied to pixels in detail regions.Meanwhile,the gain difference is larger between pixels of Retinex output associated with a small Gaussian surround space constants while the gain difference is small between pixels of Retinex output associated with a large Gaussian surround space constants.When the proposed algorithm is applied to images severely degraded by fog,experiments show that the algorithm can effectively remove fog degradation from color images.