Enhancing Underwater Images Through Non-Local Prior-Based Dehazing
Reema Verma · 2023
Underwater inspections, marine biology, and archaeology have all struggled with the same problem: poor picture quality caused by haze and colour distortion. This study explores the potential of dehazing methods based on non-local priors for improving underwater photography. Diverse underwater pictures were carefully acquired and pre-processed to illustrate a range of hazy and environmental situations. Dark channel prior, multi-scale retinex, convolutional neural networks (CNNs), and other dehazing techniques were used and analysed thoroughly. Standardised quantitative criteria, including PSNR and SSIM, confirmed the dramatic improvement in picture quality brought about by these dehazing methods. In particular, CNN-based techniques showed impressive performance, with high PSNR and SSIM values being attained. Qualitative evaluations also showed how underwater photographs were changed in appearance, with missing features being restored, contrast being boosted, and colour distortion is minimised. These results highlight the promise of dehazing methods based on non-local priors as useful resources for improving marine photography. They provide light on cutting-edge practices, which have real-world consequences for ocean exploration and study. Possible future work to better understand the ocean may entail tailoring these methods to various water conditions and incorporating them into real-time imaging systems.