Infrared Image Enhancement in Maritime Environment with
Convolutional Neural Networks
Purbaditya Bhattacharya
1
,J¨ org Riechen
2
and Udo Z ¨ olzer
1
1
Department of Signal Processing and Communications, Helmut Schmidt University, Hamburg, Germany
2
WTD 71 Military Service Center for Ships, Naval Weapons, Maritime Technology and Research, Eckernf¨ orde, Germany
Keywords: Image Processing, Convolutional Neural Network, Denoising, Super-resolution.
Abstract: Image enhancement approach with Convolutional Neural Network (CNN) for infrared (IR) images from mar-
itime environment, is proposed in this paper. The approach includes different CNNs to improve the resolution
and to reduce noise artefacts in maritime IR images. The denoising CNN employs a residual architecture
which is trained to reduce graininess and fixed pattern noise. The super-resolution CNN employs a similar ar-
chitecture to learn the mapping from a low-resolution to multi-scale high-resolution images. The performance
of the CNNs is evaluated on the IR test dataset with standard evaluation methods and the evaluation results
show an overall improvement in the quality of the IR images.
1 INTRODUCTION
Optical cameras contain sensors that are able to de-
tect light of wavelength in the range of 450 - 750 nm
and hence limited by the availability of light. Infrared
(IR) cameras and thermographic cameras in particular
have sensors that detect thermal radiation and are in-
dependent from the amount of ambient visible light.
The thermal radiation of the object determines how
salient or detailed it will be in an infrared image and
can provide useful information, otherwise not avail-
able in a normal image. IR imagery has become con-
siderably popular over the last years because of its us-
age in multiple fields of application including medical
imaging, material testing, military surveillance. Due
to its effectiveness, IR imaging is used extensively
in maritime environment for maritime safety and se-
curity application, activity detection, object tracking,
and environment monitoring.
IR images suffer from low signal-to-noise ratio
(SNR) because of the non-uniformity of the detec-
tor array responses and their underlying processing
circuits. The ambient temperature plays a very im-
portant role since the IR camera has to be calibrated
accordingly (Zhang et al., 2010). In this context,
outdoor maritime environment poses a bigger chal-
lenge compared to an indoor environment due to the
temperature fluctuations, atmospheric loss, wind, and
rain. In spite of regular camera calibrations and er-
ror correcting techniques (Zhang et al., 2010), the
(a) Original (b) Enhanced
Figure 1: An example of enhancement in IR images.
image suffers from spot noise, fixed pattern noise,
graininess, blur and other artefacts. Traditional digital
image processing techniques of image enhancement
have been extensively used over the years. The classi-
cal approaches include the usage of adaptive median
filters, gradient based approach like the total varia-
tion denoising (Micchelli et al., 2011), (Goldstein and
Osher, 2009), wavelet based approach (Zhou et al.,
2009), non-local self similarity (NSS) based methods
(Dabov et al., 2007), (J.Xu et al., 2015), and meth-
ods on sparse representation based dictionary learning
Bhattacharya, P., Riechen, J. and Zölzer, U.
Infrared Image Enhancement in Maritime Environment with Convolutional Neural Networks.
DOI: 10.5220/0006618700370046
In Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2018) - Volume 4: VISAPP, pages
37-46
ISBN: 978-989-758-290-5
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
37