Dehazing of Underwater Images Using Multiscale DehazeNet-50 Deep Learning Network

The submersible cameras contain images that contain intrinsic optical problems which include color change, contrast decrement and blurriness. The current research work is focused on enhancing the brightness as well as the sharpness of images acquired underwater, while at the same time, preventing or reducing over-contrast amplification which distorts the image structure. The research method follows preprocessing using the HSV conversion followed by threshold segmentation. Restoration of underwater images addresses issues such as haze, color shift, and low contrast with the help of our proposed Multiscale DehazeNet-50 (MDNet-50) architecture. The proposed Deep Neural Network model generalization is explained by comparing it with many models and their performance on various datasets on the ImageNet dataset. The accuracy is $96.8 \%$ which in general is higher in comparison with other kinds of deep learning studies.