Deep Learning Approach for Submerged Image Enhancement

Abstract: Because of underwater pictures application in ocean engineering, ocean research, marine biology, and marine archaeology to name a few, underwater picture enhancement was widely publicized in the last several years. Underwater photos frequently upshot in low contrast, blurred, color distortion, hazy, poor visible images. This is because of light attenuation, absorption, scattering (forward scattering and backward scattering), turbidity, floating particles. As a result, effective underwater picture solution must be developedin order to improve visibility, contrast, and color qualities for greater visual quality and optical attractiveness. Many underwater picture enhancing approaches have been proposed to overcome these challenges; however they all failed to produce accurate results. Hence for this we first undertook a large scale underwater image dataset which is trained by convolution neural network (CNN) and then we have studied and implemented a deep learning approach called very deep super resolution (VDSR) model for improving the color, contrast, and brightness of underwater photos by using different algorithms such as white balance, histogram equalization, and gamma correction respectively. Moreover, our method is compared with the existing method which reveals that our method surpassesthe existing methods Keywords: CNN, gamma correction, histogram equalization, underwater image enhancement, VDSR, white balance