激光与光电子学进展, 2019, 56 (20): 201003, 网络出版: 2019-10-22
基于全卷积回归网络的图像去雾算法 下载: 872次
Image Dehazing Algorithm Based on Full Convolution Regression Network
图像处理 图像去雾算法 卷积神经网络 端到端 限制对比度自适应直方图均衡化 image processing image dehazing algorithm convolutional neural network end-to-end contrast limit adaptive histogram equalization
摘要
针对当前去雾算法经常出现过度曝光、颜色失真等问题,提出了一种基于全卷积回归网络的去雾算法。该回归网络基于端到端系统,由特征提取和特征融合两部分构成。首先,输入有雾图像,经过特征提取和特征融合,最终回归为粗透射率图;之后使用导向滤波对其进行优化,再利用大气物理散射模型反演出无雾图像;最终采用限制对比度自适应直方图均衡化(CLAHE)对无雾图像进行增强,以得到更符合人类视觉的清晰图像。所提算法不仅可以有效避免去雾后出现的过度曝光和颜色失真等问题,而且能保留图像完整的细节信息,具有较好的去雾效果。
Abstract
Herein, a dehazing algorithm based on a full convolution regression network is proposed to solve the overexposure and color distortions caused by current dehazing algorithms. The regression network is based on an end-to-end system and comprises two parts, feature extraction and feature fusion, to which a foggy image is first subjected, then regressed into a coarse transmittance map and optimized by the guide filter. The atmospheric physical scattering model is used to invert a fog-free image , which is then enhanced via contrast limit adaptive histogram equalization (CLAHE) to obtain a clear image that is more suitable to human vision. The proposed algorithm can avoid problems such as overexposure and color distortion post dehazing, retain complete details, and provide a better dehazing effect.
张泽浩, 周卫星. 基于全卷积回归网络的图像去雾算法[J]. 激光与光电子学进展, 2019, 56(20): 201003. Zehao Zhang, Weixing Zhou. Image Dehazing Algorithm Based on Full Convolution Regression Network[J]. Laser & Optoelectronics Progress, 2019, 56(20): 201003.