Underwater image compression based on wavelet decomposition and multiscale feature enhancement

With the development of marine underwater technology, the amount of underwater image data has increased dramatically, and the limited communication bandwidth has an urgent need for high-quality compression techniques. The current image compression methods based on convolutional neural network (CNN) and generative adversarial network (GAN), when dealing with low bit rate marine underwater images, ignore the scene characteristics, resulting in texture loss and degradation of compression quality. In this paper, we propose an underwater image compression based on wavelet decomposition and multiscale feature enhancement to construct a color correction network (CCN) based on the U-Net architecture to repair image color distortion, and a multiscale wavelet transform combined with a feature enhancement module (FEM) and a channel-space dual-attention mechanism to achieve adaptive enhancement of frequency and spatial domain features, effectively preserving the key details of the image to improve the quality of the reconstructed image. Experiments show that compared with the traditional and existing algorithms, this method has better PSNR and SSIM score at low bit rates, and improves the visual quality of reconstructed images while realizing high compression ratios.