摘要
针对单一雾霾图像的去雾算法,通常以颜色增强与先行验证较为常用,目前结合深度学习应用于雾霾图像处理算法也常出现,并且取得了一定的成果和应用。本研究受两个传统的先行验证算法,即黑通道先行验证与彩度灰度值先行验证方法的启发,通过深入研究修复后雾霾图像与无雾图像RGB值的先行验证,发现其两者存在的相关性的统计规律,基于这一特点利用卷积神经网络进行学习,使得雾霾图像得到了修复,并取得了较为理想的视觉效果。后通过PSNR评价结果表明,雾霾图像修复的理论结果与目测的实践结果基本吻合,从而也证实了本研究所采用的RGB值先行验证算法具有一定的实用价值。
For the single fogging image defogging algorithm, it is usually used to color enhance and the prior. At present, it is often used in fogging image processing algorithm using combining with deep learning, then it has made some achievements and applications. This study is inspired by two traditional prior algorithms, there are dark channel prior and color attenuation prior. Through in-depth research on the prior of RGB value of repaired fogging image and non fogging image, we find the statistical law of their correlation. Based on this feature, we use convolutional neural network for deep learning, so that fogging image can be repaired, and take the ideal visual effect is obtained. Finally, the peak single to noise ratio evaluation results show that the theoretical results of fogging image restoration are basically consistent with the results of visual inspection practice, which also confirms that the RGB value prior algorithm used in this study has a certain practical value.
出处
《科技创新与应用》
2020年第1期1-5,11,共6页
Technology Innovation and Application