针对低照度图像存在亮度低、对比度低、边缘模糊等问题, 首先将低照度图像NSCT多尺度分解, 获得低频子带图像和高频子带系数; 接着将低频图像采用多尺度Retinex 提升图像亮度, 借助非线性的双边滤波函数估计其照度分量, 利用Gamma校正函数对照度分量进行校正, 提高图像的动态范围, 利用影响因子校正反射分量, 丰富其层次性, 对图像的整体轮廓进行增强; 然后将高频部分用Bayes阈值隔离噪声, 利用自适应分数阶微分对图像的边缘、纹理等细节进行增强; 最后对处理后的图像进行NSCT重构。实验结果表明,本文算法的对比度、清晰度及信息熵与现有增强方法相比, 平均提高了10.7%、9.8%、2.3%。增强后的图像在细节、边缘保持等方面也优于现有算法, 改善了图像整体的视觉效果。

Aiming to solve the low illumination images of low brightness, low contrast, and blurred edges, an NSCT (Non-Subsampled Contourlet Transform) domain images enhancement algorithm is proposed based on adaptive Retinex and adaptive fractional differential. Firstly, multi-scale NSCT decomposition of low-illuminance images is needed to obtain low-frequency subbed images and high-frequency subbed coefficients. The image brightness of low-frequency images needs to be improved by multi-scale Retinex, and using non-linear bilateral filtering function to estimate the illumination component. The illuminance component is corrected with the gamma correction functions to improve the dynamic range of the image, and the reflection component is corrected using the influence factor to enrich its hierarchy and enhance the overall outline of the image. Then, the Bayes threshold is used to isolate noise in the high-frequency part, the adaptive fractional differentiation can be used to enhance the details of the edges and textures of the image. Finally, NSCT reconstructs the processed image. The experimental results show that the contrast, sharpness and information entropy of the algorithm in this paper are improved by 10.7%, 9.8%, and 2.3% on average compared with the existing enhancement methods. The enhanced image is also superior to existing algorithms in terms of details and edge preservation, which improves the overall visual effect of the image.