dlut-dimt
commented
5 years ago The color cast of the images in this dataset is not artificially added, but is inherently. There are two main methods to add blue or green color cast to the image, one is to synthesize the image using the physical imaging model as follows: I(x)=J (x)t(x)+B (1-t(x)) where I(x), J(x), t(x) and B are synthetic underwater image, clear image, transmission map and background light, respectively. The depth map d(x) of corresponding J (x) can be obtained by existing datasets [1][2] or depth estimation methods. And t(x) can be synthesized by t(x)=exp(-βd(x)), where β=[β_r;β_g;β_b ], β_r∈[0.05,0.15], β_g∈[0.6,0.9], β_b∈[0.7,1.0], and B=[B_r;B_g;B_b ], B_r∈[0.05,0.2], B_g∈[0.5,0.8], B_b∈[0.5,1.0]. When B_g>B_b, the image is greenish, otherwise, it is bluish. Another method is synthesizing by waterGAN. For more details, please refer to the correspinding paper [3].
[1] B. Li, W. Ren, D. Fu, D. Tao, D. Feng, W. Zeng, and Z. Wang, “Benchmarking single-image dehazing and beyond,” IEEE Transactions on Image Processing, vol. 28, no. 1, pp. 492–505, 2019. [2] N. Silberman, D. Hoiem, P. Kohli, and R. Fergus, “Indoor segmentation and support inference from rgbd images,” European Conference on Computer Vision, 2012. [3] J. Li, K. A. Skinner, R. M. Eustice, and M. Johnson-Roberson, “Watergan: Unsupervised generative network to enable real-time color correction of monocular underwater images,” IEEE Robotics and Automation Letters, vol. 3, no. 1, pp. 387–394, 2018
Sir ,how to add blue or green color cast to the image .