Title: Multiple channel adjustment based on composite backbone network for underwater image enhancement
Authors: Yuhan Chen; Wende Ke; Lei Kou; Qingfeng Li; Dongxin Lu; Yan Bai; Zhen Wang; Junhe Wan
Addresses: School of System Design and Intelligent Manufacturing, Southern University of Science and Technology, Shenzhen 518055, China ' Department of Mechanical and Energy Engineering, Southern University of Science and Technology, Shenzhen 518055, China ' Institute of Oceanographic Instrumentation, Qilu University of Technology (Shandong Academy of Sciences), Qingdao 266075, China ' Health Management System Engineering Center, School of Public Health, Hangzhou Normal University, Hangzhou 311121, China ' Health Management System Engineering Center, School of Public Health, Hangzhou Normal University, Hangzhou 311121, China ' Department of Mechanical and Energy Engineering, Southern University of Science and Technology, Shenzhen 518055, China ' Department of Mechanical and Energy Engineering, Southern University of Science and Technology, Shenzhen 518055, China ' Institute of Oceanographic Instrumentation, Qilu University of Technology (Shandong Academy of Sciences), Qingdao 266075, China
Abstract: In order to heighten enhancement effects for underwater images under different conditions, multiple channel adjustment based on composite backbone network (MC-CBNet) was proposed that skilfully combine the enhancement effect from RGB colour space as well as HSV and Lab. MC-CBNet consists of a preliminary enhance block, a multi-space adjust block and a confidence map block. Preliminary enhance block and multi-space adjust block adjust the images from RGB, HSV and Lab colour spaces respectively. The confidence map is obtained by the ultimate block to fuse the results of different channels. Besides, preliminary enhance block and confidence map block are formed from composite backbones. Experimental results on underwater image enhancement benchmark (UIEB) indicate that our method gets better grades than existing methods under both reference subset and challenging subset evaluation.
Keywords: underwater image; enhancement; deep learning; composite backbone; multiple channel.
DOI: 10.1504/IJBIC.2023.135476
International Journal of Bio-Inspired Computation, 2023 Vol.22 No.3, pp.162 - 175
Received: 21 Oct 2022
Accepted: 22 Jul 2023
Published online: 14 Dec 2023 *