Physical model based deep learning method for underwater image restoration
A deep learning method based on physical model for underwater image restoration was proposed to address color distortion,detail blurring,and contrast degradation of the images in complex underwater environments.The physical model of underwater optical imaging was used to constrain the generative adversarial network(GAN),adding theoretical support to the deep learning method while lowering reliance of the deep learning restoration effect on the training dataset.Initially,crucial parameters of the physical model were estimated,simplifying traditional methods for computing transmission maps based on visual saliency principles.Then,the haze map obtained by the generator of GAN was used to calculate ambient light.Subsequently,the underwater restoration images were computed through the physical model,and the discriminator of GAN was used to distinguish between the restoration and reference images.Through iterative optimization of network parameters,underwater image restoration was achieved.Experimental results on the test set show that the image restoration effect has been effectively improved in various subjective and objective evaluation indicators.
underwater image restorationdeep learningphysical model of underwater imaging
万方数据
维普
