基于颜色衰减先验的小波融合图像去雾方法

发布时间：2018-03-11 07:01

本文选题：图像去雾　切入点：颜色衰减先验　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：人类的生产生活对自然环境造成了严重影响,导致雾霾天气出现的日渐频繁。雾天气条件下拍摄的照片模糊不清,对各类视频监控系统的使用带来了不便,因此对图像进行去雾处理,恢复出清晰的图像,是现在的研究热点,也是很有实际价值的课题。根据颜色衰减先验知识进行去雾是近两年提出的一个新的去雾方法,该方法快速有效,极大的提高了去雾速度,具有推广价值。笔者在实验过程中发现,该算法中大气散射系数的选择对最终的去雾结果影响很大,需要进行人工调整。而且,颜色衰减先验知识,在雾浓的区域失效,因此对雾浓的图像处理效果不佳。本文针对这两个问题进行了改进。本文在颜色衰减先验知识的基础上,提出了基于小波融合的图像去雾算法。首先,建立了透射率关于图像亮度、饱和度的线性模型。其次,选用400幅图像及其准确的景深信息用于训练样本的生成,并将大气散射系数的概率分布直接融入到训练样本中,保证了训练样本生成的准确性,随后采用机器学习中的监督学习算法估计出图像透射率,然后用小波融合算法将估计出的透射率与图像灰度图的反转图像融合。最后,用细化后的透射率信息对图像进行去雾处理。直接对透射率进行建模的过程,避免了大气散射系数的选择,用图像灰度图的反转图对透射率进行细化,提高了透射率的准确性,最终提高了对雾浓图像的处理效果。实验结果证明本文算法可行有效。此外文章采用通用的去雾图像质量评价方法将本算法与先进的去雾算法进行了比较。实验结果表明,本文算法提高了透射率的准确性,改善了浓雾区域的恢复效果。具有适用性好,计算复杂度低的特点。本文的主要贡献为以下几点:(1)建立了透射率关于图像亮度饱和度的线性模型,避免了去雾过程中对大气散射系数的人工选择。(2)得到大气散射系数的分布直方图,并将其代入训练样本中,从而获得准确度高的训练样本,保证训练出模型的可靠性。(3)提出图像灰度图的反转图能够作为透射率对图像进行去雾处理,尤其对雾浓的图像,处理效果较佳。(4)利用小波算法将估计出的透射率与图像灰度图的反转图融合,提高透射率的准确性,保证最终的去雾效果。
[Abstract]:The production and life of human beings have caused a serious impact on the natural environment, resulting in the increasing frequency of haze weather. The photographs taken under the fog weather conditions are blurred, which brings inconvenience to the use of all kinds of video surveillance systems. Therefore, to defog the image and restore the clear image is a hot research topic and a very valuable subject. According to the prior knowledge of color attenuation, de-fogging is a new de-fogging method proposed in the last two years. This method is fast and effective, greatly improves the speed of de-fogging, and is worth popularizing. In the course of experiment, the author finds that the selection of atmospheric scattering coefficient in this algorithm has a great influence on the final de-fogging result and needs to be adjusted manually. The prior knowledge of color attenuation is invalid in the region of fog concentration, so the image processing effect of fog concentration is not good. This paper improves these two problems, and based on the prior knowledge of color attenuation, An image de-fogging algorithm based on wavelet fusion is proposed. Firstly, a linear model of transmittance for image brightness and saturation is established. Secondly, 400 images and their accurate depth of field information are used to generate training samples. The probability distribution of atmospheric scattering coefficient is directly incorporated into the training sample to ensure the accuracy of the training sample generation. Then the image transmittance is estimated by the supervised learning algorithm in machine learning. Then wavelet fusion algorithm is used to fuse the estimated transmittance with the inverse image of the gray image. Finally, the thinned transmittance information is used to defog the image. The atmospheric scattering coefficient is avoided, and the transmissivity is thinned by the inverse image of the image grayscale image, which improves the accuracy of the transmittance. Finally, the processing effect of fog concentration image is improved. The experimental results show that the proposed algorithm is feasible and effective. In addition, this algorithm is compared with the advanced de-fogging algorithm by using the general evaluation method of de-fogging image quality. The experimental results show that, The algorithm improves the accuracy of transmittance and improves the recovery effect of dense fog region. The main contributions of this paper are as follows: 1) A linear model of transmittance for image luminance saturation is established. The distribution histogram of atmospheric scattering coefficient is obtained by avoiding the artificial selection of atmospheric scattering coefficient in the de-fogging process, and the distribution histogram of atmospheric scattering coefficient is added to the training sample to obtain the training sample with high accuracy. To ensure the reliability of the model, it is proposed that the inverse image of gray image can be used as transmittance to defog the image, especially for the dense fog image. The wavelet algorithm is used to fuse the estimated transmittance with the inverse image of gray image to improve the accuracy of transmittance and ensure the final effect of fog removal.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】