Single Underwater Image Restoration By Multi-Scale Fusion In Multi-Color Space

Abstract

Underwater imaging technology is a key technology for people to explore the world under the sea surface. However, the different attenuation of lights with different wavelengths and the scattering of light by suspended particles in water seriously affect the quality of underwater images. Most of the previous researches focused on eliminating blur solely or color cast solely in the image or these methods need a mass of image databases. To break through the limitations of hardware and the large image databases, we want to produce a new method that can be applied to more underwater scenes and can eliminate the image blur and color cast simultaneously. We propose a multi-scale fusion in multi-color space (MFMC) method, firstly, we obtain two images to be fused from one underwater image, one from stretching and shrinking the original underwater image in multiple color Spaces, and the other one from defogging the original underwater image based on the underwater imaging model. Secondly, we obtain four weight images from two images to be fused and then normalize the four weight images into two weight images. Thirdly, we decompose two images that are to be fused and two weight images into Laplace image pyramids and Gaussian image pyramids with the same level respectively. Finally, we fuse these pyramids and upsample the top image of the fusion image pyramid until the upsampled image has the same size as the original underwater image, and the final upsampled image is restored underwater image. we discuss two classical methods to eliminate blurring and color cast and an up to date underwater image restoration method. And we compare their performance with that of MFMC on three classic metrics and one up to date metric. Qualitative and quantitative analysss prove that our method was significantly effective in the restoration of underwater images.