基于人眼亮度阈值特性的图像增强算法  

Human vision system brightness threshold based image enhancement method

作　　者：范晓鹏[1,2] 蔡铁峰[1,2] 朱枫[1] 郝颖明[1] 吴清潇[1] 夏仁波[1] 

机构地区：[1]中国科学院沈阳自动化研究所中国科学院光电信息处理重点实验室辽宁省图像理解与视觉计算重点实验室,辽宁沈阳110016 [2]中国科学院大学,北京100049

出　　处：《光电子．激光》2014年第8期1606-1612,共7页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金项目(51005229)资助项目

摘　　要：结合人眼视觉特性,提出一种面向人眼视觉的图像增强算法。在保持原图像所有像素灰度大小关系不发生倒序的前提下,以人眼可分辨的像素对数作为目标函数。为使目标函数实现最大化,首先根据人眼亮度阈值特性测试指定显示器的人眼最多可分辨的灰度级数;然后结合图像本身的灰度级数与人眼最多可分辨的灰度级数使用一种动态规划灰度级合并算法,使得灰度合并后图像中包含的灰度级数不多于人眼最多可分辨灰度级数;最后对合并后的灰度按人眼视觉亮度阈值特性做一一映射。本文算法充分考虑了图像观测过程中人眼感知与显示器显示亮度,具有明确的目标函数,图像增强效果优于目前常用算法。Combined with the human visual characteristics, we propose an image enhancement algorithm for the human eye. Objective function of the proposed algorithm is pixel pairsr number, of which the gray-scale difference can be perceived by human eye. In order to keep the fidelity of image information, the enhancement process should be under the condition of maintaining the gray scale difference direction of the original image pixels. To maximize the objective function,the first step is to test the human eye maximum distinguishable gray levels towards specified monitor. The test should consider human eye brightness threshold characteristics. Combined with the gray levels of the image itself and the maximum gray levels human eye can distinguish,we use a dynamic programming algorithm to compress gray lev- els,so that the gray levels contained in the merged image would not be more than the maximum gray levels human eye can distinguish. Finally, the grays implement one to one mapping based on human visual brightness threshold characteristics. Compared with the novel image enhancement algorithms, the pro- posed algorithm is fully considering the interaction of human eye and the monitor brightness. Further more, the proposed algorithm has a clear objective function. Experimental results show that the proposed algorithm has better performance than the currently used image enhancement algorithms.

关 键 词：图像增强 目标函数 人眼亮度阈值 恰可见偏差(JND) 最多可分辨灰度级数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]