基于大气散射模型的去雾算法研究-计算机应用技术专业论文.docxVIP

摘 要 计算机视觉系统获取的图像质量直接影响其实用价值。实际环境中，因雾、 霾、雨、雪等恶劣天气存在，室外视觉系统获取的图像质量会大大降低。本文对 基于大气散射模型的单幅图像去雾算法进行研究，分别对 fattal 和何凯明两位作 者提出的两种代表性算法进行改进，以获得更佳去雾效果。 基于大气散射模型的图像去雾算法需要解决两个问题：估计大气光 A 与估计 透射率 t。针对 fattal 去雾模型，重点对估计大气光与估计透射率 t 做了深入分析， 采用基于颜色及纹理相似度的图像分割算法获得区域分割图，计算透射率，根据 大气散射模型恢复含雾图像。针对该方法出现的小区域偏色问题，采用基于小波 的后处理方法，实验表明该方法效果较好。 现有方法对浓雾且与背景与大气相似时的图片去雾效果较差，且暗通道去雾 存在恢复图像偏暗、对比度不够、图像存在光晕和细节不突出等问题，本文提出 一种改进的中值暗通道优先的图像去雾算法。新算法将去雾图像的前处理与后处 理结合。暗通道去雾前，采用二维模态分解的方法对含雾图像进行预处理，利用 暗通道优先原理计算全局大气光，获得随深度变化的大气传输衰减过程，并对其 进行双边滤波和带色彩恢复的多尺度视网膜增强（MSRCR），以达到保留图像细 节信息和提高图像局部对比度的目的，得到场景的反射图像，实现图像去雾。实 验表明，该方法不仅能消除图像去雾后产生的“光晕”现象，进一步丰富图像细节 信息，且能在保持图像色彩信息的基础上，动态地提高图像对比度，达到较为理 想的去雾效果。 关键词: 图像去雾 暗通道 视网膜增强 二维模态分解 双边滤波 ABSTRACT The quality of image obtained by computer vision can directly affect the practical value. In the real environment, The quality of image obtained by outdoor computer vision will be reduced by the existence of snow, rain, fog, and haze, etc. This paper studied the Single image haze removal technology based on atmosphere physical model. Improved the typical methods proposed by Fattals and HE KAIMING, aiming to obtain the better dehaze results. The method based on atmosphere physical model, need to solve two essential problem: estimate airlight-A and transmittance-t, Focused on fattal method to analyze the estimation of airlight and transmittance.We use image color and texture similarity segmentation algorithm to obtain segmentation map and Calculation of the transmittance.Then Complete dehazing,according to the atmospheric model. In order to solve the problem of partial color,then enhance the image based on the method of wavelet,experiment shows that this method is quite practical. A new image dehaze algorithm based on the median dark channel prior is proposed to deal with the halos and no highlighting of details what the images after using the existing dehaze methods. At first, use dark channel prior to calculate the global atmospheric light, and get the atmospheric transmission