Underwater vision suffers from severe effects due to selective attenuationand scattering when light propagates through water. Such degradation not onlyaffects the quality of underwater images but limits the ability of visiontasks. Different from existing methods which either ignore the wavelengthdependency of the attenuation or assume a specific spectral profile, we tacklecolor distortion problem of underwater image from a new view. In this letter,we propose a weakly supervised color transfer method to correct colordistortion, which relaxes the need of paired underwater images for training andallows for the underwater images unknown where were taken. Inspired byCycle-Consistent Adversarial Networks, we design a multi-term loss functionincluding adversarial loss, cycle consistency loss, and SSIM (StructuralSimilarity Index Measure) loss, which allows the content and structure of thecorrected result the same as the input, but the color as if the image was takenwithout the water. Experiments on underwater images captured under diversescenes show that our method produces visually pleasing results, evenoutperforms the art-of-the-state methods. Besides, our method can improve theperformance of vision tasks.
展开▼