A Real-time and Unsupervised Advancement Scheme for Underwater Machine Vision
Xingyu Chen, Zhengxing Wu, Junzhi Yu, Li Wen · 2017
This paper presents a real-time and unsupervised advancement scheme (RUAS) for underwater machine vision in the natural light condition. RUAS consists of three steps, pre-searching, restoration, and post-enhancing. In pre-searching, we provide a Protected and Greedy Artificial Fish School Algorithm (PGAFSA) to optimize the key parameters of the underwater images, and design an evaluating indicator for the PGAFSA based on the features of underwater images. During the restoration, an image degeneration model is built and the Wiener Filter is employed for noise suppression. Moreover, a filtering-aided color correlation method (FCCM) is then presented against color absorption caused by water. The contrast limited adaptive histogram equalization is employed for the contrast stretch in post-enhancing. Finally, we validated the effectiveness and feasibility of the proposed RUAS with deep-sea environmental videos and practical underwater environments.