摘要
针对目前单幅雾霾降质图像存在大面积天空域,导致暗原色失效复原图像失真以及去雾时间复杂度高的问题。提出一种基于暗原色先验和快速引导滤波的去雾方法,针对存在高亮天空域图像多尺度滤波采用自适应阈值分割得到天空域,在天空域求取精确大气光值,再将天空域和非天空域透射率有效归一化,最后采用暗通道图作为引导图快速引导滤波精细化透射率,最大限度保留边缘细节的同时有效降低时间复杂度。实验结果表明,该方法可以有效处理天空域,保持色彩和细节信息,在算法实时性上有明显优势。
Aim at the problem of single fog haze drop quality image exists a big area airspace, which led to dark channel prior fail- ure to dehaze image and a very high-time-complexity. To solve these problem, a new algorithm of haze removal based on dark channel prior and fast guided filter is proposed. First in the highlighted sky image used adapted threshold segmentation divide the sky region and nonsky, then take atmosphere light in this region, and make the sky and non-sky domain transmission rate effective return. And first use the dark channel raher than the color hazy image as the guidance images,to optimize the transmission, which can achieve rather good dehazing results, but with a relative simple implementation and low-time-complexity. Experimental results indicate that this method can effectively deal with the sky domain, keep the color and detail information and has a clear advantage in the real-time.
出处
《电视技术》
北大核心
2016年第6期22-27,38,共7页
Video Engineering
基金
重庆市研究生科研创新基金项目(CYS15166)
关键词
暗原色先验
阈值分割
快速引导滤波
dark channel prior
sky region segmentation
fast guided filter