基于环境光物理模型的快速图像清晰化技术研究

发布时间：2018-03-18 16:43

本文选题：暗通道先验　切入点：边缘检测　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：天空的悬浮粒子能散射和吸收光线形成雾霾,造成室外拍摄成像的对比度和保真度都较低。在室外图像质量要求较高的应用,如军事敌情监控,刑侦犯罪细节斟酌,汽车前方障碍物观测等,都需要对降质图像做处理,还原目标细节。对雾霾清晰化处理能显著提高图像质量,使之更好地应用于高级图像处理领域。论文针对雾霾条件下单幅图像降质以及现有去雾方法时间复杂度高的问题,以环境光物理模型为基础,在暗通道去雾的基础上,引出快速视觉优化去雾算法。设计和实现了两种获得大气光向量A的方案,其一:结合LOG边缘检测和四叉树快速定位大气光向量A的区间,并应用skyline算法获取A值;其二:对单幅图像阈值分割找到天空区域,进而选用二分图循环模型定位精确的大气光向量A。然后针对暗通道先验模型确定的初始透射率t(x,y)采用改进的最小二乘方滤波进行优化,可保证其边缘细节完整且受到噪声的影响程度小,同时抑制晕轮效应,之后通过膨胀和腐蚀方法进一步优化投射图的边缘。最后利用求得的大气光向量A和优化后的透射率参量,代入环境光模型还原清晰无雾的图像。使用大量测试图像对本文的算法加以测试,从客观和主观两个角度对本文算法进行评估,实验结果表明,改进算法在保障优化去雾视觉效果的前提下节约处理时间;梯度,信息熵和视觉保真度综合指标均优于对比算法。
[Abstract]:Suspended particles in the sky can scatter and absorb light to form haze, resulting in low contrast and fidelity of outdoor imaging. In view of obstacles in front of the vehicle, it is necessary to process the degraded image and restore the target details. The clear processing of haze can significantly improve the image quality. Aiming at the problem of single image degradation under haze condition and the high time complexity of existing de-fogging methods, based on the environmental photophysical model and the dark channel de-fogging, this paper aims at solving the problem that it is better applied to the field of advanced image processing. Two schemes of obtaining atmospheric light vector A are designed and implemented. First, combining LOG edge detection and quadtree fast locating the interval of atmospheric light vector A, and using skyline algorithm to obtain A value; Second: find the sky region for the threshold segmentation of a single image, Then the bipartite graph cycle model is used to locate the accurate atmospheric light vector A. then the improved least square filter is used to optimize the initial transmissivity determined by the dark channel priori model. The edge details can be guaranteed to be complete and less affected by noise, and the halo effect can be suppressed at the same time. The edge of the projection map is further optimized by means of expansion and corrosion. Finally, the obtained atmospheric light vector A and the optimized transmittance parameter are used. A large number of test images are used to test the algorithm in this paper, and the algorithm is evaluated from both objective and subjective perspectives. The experimental results show that, The improved algorithm saves processing time on the premise of optimizing the visual effect of de-fogging, and the comprehensive indexes of gradient, information entropy and visual fidelity are better than the contrast algorithm.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】