Underwater Image Enhancement using a Light Convolutional Neural Network and 2D Histogram Equalization

Abstract

Underwater images usually have low contrast, blurring, and extreme color distortion because the light is refracted, scattered, and absorbed as it passes through the water. These features can lead to challenges in image-based processing and analysis. In this paper, a two-step method based on a deep convolution network is proposed for solving these problems and enhancing underwater images. First, through a light global-local structure, the initial image enhancement is performed and the color distortion and degradation of the images are partially covered. Then, two-dimensional histogram equalization is used as an appropriate complement to the network. Two-dimensional histogram equalizing is able to produce clear results and prevents excessive contrast. The results show that the proposed method performs better than other methods in this field in terms of qualitative and quantitative criteria.