基于Retinex亮度校正与感知对比度的图像增强算法

Image enhancement algorithm based on Retinex luminance correction and perceptual contrast

针对当前图像增强中易出现光晕伪影、细节裁剪效应和过度增强等问题,基于人类视觉系统(human visual system,HVS),优化对复杂与不同照明条件中场景感知力,提出了一种有效的非均匀与低光照图像的自然增强算法。首先,基于白块假设估计输入图像的亮度,并定义一种自适应的Retinex亮度校正算子,改善图像亮度。对于明亮区域的对比度,提出了一种调整映射函数,通过添加细节增益因子,结合调整映射函数与增益因子,保护与改善明亮区域的细节与对比度,有效抑制光晕、裁剪效应。对于暗淡区域,构建改进高斯差分(difference of Gaussian,Do G)的感知对比度映射(perceptual contrast map,PCM),通过动态范围调整的PCM来调节对比度,实现感知对比度增强与噪声抑制,防止过度增强。最后,采用一个显著图引导的加权组合算法,将暗区域对比度增强结果与明亮区域细节增强结果进行加权融合,并利用色彩校正技术来恢复图像的色彩信息,从而输出最终的增强结果。实验结果表明,与当前图像增强技术相比,所提方法的增强图像具有更好的视觉质量和感知对比度,以及更为清晰的细节与色彩信息。

In order to solve the problems such as halo artifact,detail cutting effect and over-enhancement in the current image enhancement,an effective natural enhancement algorithm for non-homogeneous and low illumination image based on the human vision system was proposed to optimize the scene perception in complex and different lighting conditions. First of all,white luminance of the input image was estimated based on assumptions,and an adaptive Retinex operator was defined to improve the brightness of the image brightness correction. Secondly,an adjustment mapping function was proposed for the contrast of bright areas. By adding details gain factor and adjusting mapping function and gain factor,the detail and contrast of bright area could be protected and improved,and the halo and the clipping effect were suppressed. Then,for the dark area,the perceptual contrast mapping( perceptual contrast map,PCM)of improved Gauss differential( difference of Gaussian,Do G) was constructed,the contrast was adjusted by the dynamic range adjustment of PCM,the perceived contrast enhancement and the noise suppression were achieved to prevent excessive enhancement.Finally,a weighted combination algorithm based on a simplified saliency map was introduced to fuse the dark area contrast enhancement results with the bright area detail enhancement results,and the color correction was used to restore the color richness,in order to get the final enhancement results. The experimental results show that compared with many advanced algorithms,the image quality and perceptual contrast generated by this method are better,the details are clear,and the color is rich and with clearer details and color information.