Single Enhancement Techniques for Underwater Stereo Vision

Abstract

In order to ensure safety and efficiency of unmanned underwater works underwater situation visualization is an essential technique among unmanned automatic technologies. Although optical color images are used widely for underwater visualization, physical distortions such as color casting and blurring are occurred in underwater color images and it leads to uncertainty in depth estimation for stereo vision. In this study, we applied image fusion and deep learning techniques to improve the underwater image quality. The performance of single image enhancement techniques was evaluated by underwater image quality measure (UIQM) and underwater color image quality evaluation (UCIQE). Single image enhancement increased quantitatively UIQM and UCIQE by 82.8 % and 35.8 % on the average, respectively and improved the color accuracies of reconstructed red, green and blue – depth (RGB-D) point clouds compared to the original images. Single underwater image enhancement techniques improved the accuracies of depth estimation for underwater stereo vision.