Format

Send to

Choose Destination
Sensors (Basel). 2018 Jan 12;18(1). pii: E196. doi: 10.3390/s18010196.

Underwater Object Segmentation Based on Optical Features.

Author information

1
College of Computer and Information, Hohai University, Nanjing 211100, China. chenzhe@hhu.edu.cn.
2
College of Computer and Information, Hohai University, Nanjing 211100, China. zz_hhuc@hhu.edu.cn.
3
Laboratory of Information Optics and Opto-Electronic Technology, Shanghai Institute of Optics and Fine Mechanics, Shanghai 201800, China. buyang@siom.ac.cn.
4
Laboratory of Information Optics and Opto-Electronic Technology, Shanghai Institute of Optics and Fine Mechanics, Shanghai 201800, China. fzdai@siom.ac.cn.
5
School of Information Engineering, Nanchang Institute of Technology, Nanchang 330099, China. fantanghuai@nit.edu.cn.
6
College of Computer and Information, Hohai University, Nanjing 211100, China. hbwang@hhu.edu.cn.

Abstract

Underwater optical environments are seriously affected by various optical inputs, such as artificial light, sky light, and ambient scattered light. The latter two can block underwater object segmentation tasks, since they inhibit the emergence of objects of interest and distort image information, while artificial light can contribute to segmentation. Artificial light often focuses on the object of interest, and, therefore, we can initially identify the region of target objects if the collimation of artificial light is recognized. Based on this concept, we propose an optical feature extraction, calculation, and decision method to identify the collimated region of artificial light as a candidate object region. Then, the second phase employs a level set method to segment the objects of interest within the candidate region. This two-phase structure largely removes background noise and highlights the outline of underwater objects. We test the performance of the method with diverse underwater datasets, demonstrating that it outperforms previous methods.

KEYWORDS:

artificial light guidance; level-set-based object segmentation; optical features; underwater object segmentation

Supplemental Content

Full text links

Icon for Multidisciplinary Digital Publishing Institute (MDPI) Icon for PubMed Central
Loading ...
Support Center