激光与光电子学进展, 2019, 56 (3): 031008, 网络出版: 2019-07-31
基于场景深度估计和白平衡的水下图像复原 下载: 1834次
Underwater Image Restoration Method Based on Scene Depth Estimation and White Balance
图像处理 水下图像复原 场景深度估计 透射率估计 白平衡 颜色校正 image processing underwater image restoration scene depth estimation transmission estimation white balance algorithm color correction
摘要
由于水体对光的吸收和散射作用,水下图像往往存在对比度低、细节模糊和颜色失真等问题,为此提出一种基于场景深度估计和白平衡的水下图像复原方法。首先,采用Sobel边缘检测和形态学闭运算将水下图像中与场景深度相关的块分离,对RGB通道与场景深度的变化关系进行回归分析,以得到场景深度图像并估计水下背景光;其次,对衰减严重的颜色通道取其逆通道,以修正透射率估计;然后,通过逆求解水下光学成像模型来消除后向散射;最后,改进白平衡算法以更好地校正水下图像的颜色畸变,得到复原后的水下图像。与典型的4种水下图像复原方法进行主客观评价比较,实验结果表明,该方法可以有效地提升低质量、低照度的水下图像的细节清晰度和色彩保真度,恢复真实的视觉效果。
Abstract
Due to the absorption and scattering of the light, underwater images often have the problems of low contrast, detail blurring and color distortion. An underwater image restoration method based on scene depth estimation and white balance is proposed. First, the underwater image is divided to the scene depth related patches by using Sobel edge detection and morphological closed operation. Regression analysis of the relationship between the RGB channel and the scene depth is carried out to obtain scene depth images and to estimate the background light underwater. Second, for the heavily attenuated color channel, its inverse channel is used for correcting the transmission estimation. Third, the backscattering is eliminated by inversely calculating the underwater optical imaging model. Finally, the improved white balance algorithm is utilized to achieve better color correction of underwater images and the restored underwater images are obtained. Compared with four kinds of typical underwater image restoration methods, the experimental results demonstrate that the proposed method can effectively improve the detail clarity and color fidelity of underwater images with low quality and low illumination, while restoring the natural visual effects.
蔡晨东, 霍冠英, 周妍, 韩辉. 基于场景深度估计和白平衡的水下图像复原[J]. 激光与光电子学进展, 2019, 56(3): 031008. Chendong Cai, Guanying Huo, Yan Zhou, Hui Han. Underwater Image Restoration Method Based on Scene Depth Estimation and White Balance[J]. Laser & Optoelectronics Progress, 2019, 56(3): 031008.