基于天空区域分割的快速去雾算法研究

Research on fast dehazing algorithm based on sky region segmentation

针对传统暗通道先验去雾算法在处理户外含雾图像时,出现天空区域颜色失真和处理速度慢的问题,提出一种可自适应识别天空区域的快速去雾算法。在天空分割方面,选用图像中细节特点明显的亮度分量为研究对象,结合最大类间方差法(OTSU)和动态参数建立自适应识别天空区域算法模型,得到最佳分割阈值,分割出有雾图像的天空区域和非天空区域,并根据天空区域计算出大气光值。在提高处理速度方面,在使用引导滤波优化透射率过程中引入图像下采样算法,保证复原后图像质量的同时减少算法耗时。最后经过与多种经典算法对比,文中算法在视觉效果上细节更加自然,SSIM、PSNR和MSE的综合指标均超过其他算法,并且处理速度较快。主观和客观评价结果均表明,文中算法在视觉效果和时间效率方面都优于其他几种算法,具有一定的实用价值。

In view of the color distortion and slow processing speed in the region of sky when the traditional dark channel prior(DCP)dehazing algorithm is used to process outdoor foggy images,a fast dehazing algorithm which can identify the region of sky adaptively is proposed.In terms of sky region segmentation,the luminance components with prominent details in the image are chosen as the subject of study.In combination with the OTSU algorithm and the dynamic parameters,an algorithm model is established to identify the region of sky adaptively,and the optimal segmentation threshold value is obtained.The regions of sky and non⁃sky in the foggy images are segmented,and the atmospheric light value is computed according to the region of sky.In terms of increasing the processing speed,the image downsampling algorithm is introduced in the process of optimizing the trans⁃mittance by guided filtering,which reduces the time consumption of the algorithm while ensuring image quality after restoration.In comparison with various classical algorithms,the proposed algorithm offers more natural visual details.Its SSIM(structure similarity index measure),PSNR(peak signal⁃to⁃noise ratio)and MSE(mean squared error)surpass those of the other algorithms,and its processing speed is faster.Both subjective and objective evaluation results show that the proposed algorithm is superior to the other algorithms in visual effect and time efficiency.To sum up,it has a certain practical value.