改进的多尺度Retinex图像增强算法

针对多尺度Retinex算法在图像增强的过程中，存在着算法运算量大的问题，提出了一种基于快速二维卷积和多尺度连续估计的算法。该算法充分利用二维图像高斯卷积的可分离性和多尺度照射光连续估计的可行性，降低了Retinex算法的复杂度。同时对于增强后图像色彩容易失真的现象，提出了一种去极值的直方图裁剪法，用于保持图像色彩信息和提高对比度。实验结果表明，这些改进可以有效提高算法运行速度和改善图像增强效果。

Improved multi-scale retinex image enhancement algorithm

For the existing problems of expensive calculation in the process of image enhancement with multi-scale Retinex, this paper presented an improved algorithm on the basis of fast two-dimensional convolution and multi-scale continuous estimation. Compared with the original algorithm, the new one was simpler for its superiority of making full use of the separability of the image Gaussian convolution and the feasibility of multi-scale continuous illumination estimation. Meanwhile, a histogram clipping method with removing extremum was proposed to deal with the color distortion of enhanced image, which could restore image color information and enhance image contrast. The experimental results demonstrate that the proposed method can efficiently increase the running speed while improving image quality.