Underwater Imagery Enhancement with Multi-Channel Histogram Equalization, Depth-Adaptive Correction, and Deep Reinforcement Learning

Abstract

Underwater imagery plays a pivotal role in various scientific, industrial, and recreational applications, ranging from marine biology and oceanography to underwater archaeology and resource exploration. However, capturing high-quality images in underwater environments poses unique challenges due to factors such as light attenuation, color distortion, and particulate matter suspended in the water. In recent years, significant advancements have been made in the development of advanced techniques for enhancing underwater imagery aimed at improving perceptual quality, sharpness, and detail preservation. This research explores state-of-the-art enhancement techniques, focusing on multi-channel histogram equalization and depth-adaptive correction in the context of deep reinforcement learning (DRL) and variational autoencoders (VAE). The research presents a comprehensive analysis of the performance of these techniques using numerical metrics such as perceptual quality, sharpness, and detail preservation, derived from experimental evaluations. The findings demonstrate the effectiveness of VAE in enhancing perceptual quality and sharpness, with underwater image dehazing exhibiting commendable results as well. Furthermore, Convolutional Neural Networks (CNN) emerge as a promising approach for detail preservation in underwater images. By advancing the capabilities of underwater image enhancement techniques, this research contributes to the broader goal of unlocking the full potential of underwater exploration and understanding. The findings reveal significant improvements in image quality achieved by these techniques. VAE exhibits the highest perceptual quality scores, averaging around 0.84, with sharpness scores reaching approximately 0.91.