
Restoration of images degraded by underwater turbulence using structure tensor oriented image quality (STOIQ) metric
Author(s) -
Andrey Kanaev,
Weilin Hou,
Sergio R. Restaino,
Silvia Matt,
Szymon Gładysz
Publication year - 2015
Publication title -
optics express
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.394
H-Index - 271
ISSN - 1094-4087
DOI - 10.1364/oe.23.017077
Subject(s) - underwater , image restoration , image quality , distortion (music) , metric (unit) , computer science , computer vision , artificial intelligence , image processing , structure tensor , noise (video) , optics , physics , image (mathematics) , geology , engineering , telecommunications , amplifier , oceanography , operations management , bandwidth (computing)
Recent advances in image processing for atmospheric propagation have provided a foundation for tackling the similar but perhaps more complex problem of underwater imaging, which is impaired by scattering and optical turbulence. As a result of these impairments underwater imagery suffers from excessive noise, blur, and distortion. Underwater turbulence impact on light propagation becomes critical at longer distances as well as near thermocline and mixing layers. In this work, we demonstrate a method for restoration of underwater images that are severely degraded by underwater turbulence. The key element of the approach is derivation of a structure tensor oriented image quality metric, which is subsequently incorporated into a lucky patch image processing framework. The utility of the proposed image quality measure guided by local edge strength and orientation is emphasized by comparing the restoration results to an unsuccessful restoration obtained with equivalent processing utilizing a standard isotropic metric. Advantages of the proposed approach versus three other state-of-the-art image restoration techniques are demonstrated using the data obtained in the laboratory water tank and in a natural environment underwater experiment. Quantitative comparison of the restoration results is performed via structural similarity index measure and normalized mutual information metric.