Underwater image enhancement based on improved generative adversarial network
Aiming at the problems of color distortion and blurred details of underwater images,an underwater image enhancement algorithm based on improved generative adversarial network is proposed.The method uses the generative adversarial network as the basic structure,the generative network adopts the coding and decoding structure,and introduces the RGB color space block,the HSV color space block and the attention mechanism;the RGB block can better denoise and remove the color cast,and the HSV color space can adjust the brightness,color and saturation of the underwater image,and finally the generative network generates the image by assigning weights.The discriminant network adopts a structure similar to the Markov discriminator.Furthermore,by constructing global similarity and content-aware multinomial loss functions,the generated images are made consistent with reference images in terms of color,content and structure.Experiments show that the proposed method performs well on both subjective comparisons and objective metrics.Among them,the average values of structural similarity,peak signal-to-noise ratio,underwater color quality assessment and underwater image quality metrics in the synthetic underwater image test set are 0.774 6,19.275 8,0.488 9 and 3.312 4.The average values on the test set of real underwater images are 0.900 0,24.263 6,0.449 9 and 3.161 9.In terms of subjective evaluation and objective evaluation indicators,generally speaking,the experimental results of the algorithm in this paper are better than those of the comparison algorithm.
NETL
NSTL
万方数据
维普
