Underwater image enhancement algorithm based on dense features

Abstract

Due to the optical transmission characteristics, underwater images are degraded to different degrees than natural images, and there are many problems such as color deviation, poor visibility and fuzzy details. Based on the generation of general adverse networks (GAN), an algorithm of underwater image enhancement dense fusion GAN(DFGAN) algorithm was proposed. The algorithm improved the existing GAN and corrected the blue and green bias of the image. Meanwhile, this paper constructed a dense feature fusion mechanism in the generator, which made up for the loss of detail information in the process of image correction by making full use of the feature information in the image. The experimental results show that the enhanced image of this algorithm has been improved in the numerical value of four general objective quality evaluation indexes, such as structural similarity, peak signal-to-noise ratio, information entropy and underwater color image quality evaluation. It is proved that the algorithm can effectively correct the color of underwater low-quality image, recover texture characteristics and improve visual quality.

Keywords

• underwater image
• generative adversarial networks
• color correction
• feature fusion
• image enhancement