Processing of Hyperspectral Medical Images Applications in

Dermatology Using Matlab®

Visit the link below to download the full version of this book:

https://medidownload.com/product/processing-of-hyperspectral-medical-images-appl
ications-in-dermatology-using-matlab/

Click Download Now

Foreword

Medical imaging is a field of knowledge dealing with the methods of acquisition

and analysis of images occurring in biological and medical research. The acquired
images are used for research, diagnostic, therapeutic or educational purposes. The
rapid development of diagnostic medical equipment and information technology
enables the growing interaction of these two areas of expertise for the benefit of
patients.
It is generally difficult to show the characteristics of real medical images pic-
torially and in a useful form for a physician. In addition to qualitative assessment,
the physician also needs quantification of medical images, which will illustrate the
various diagnostic parameters of medical objects. On their basis, the physician
makes decisions related to the course of treatment, the strategy and the selection of
appropriate drugs.
Quantitative assessment, achievable through the analysis of biomedical images,
involves profiling contemporary analysis methods and algorithms. Such methods
include not only image filtering but also its morphological and point transforma-
tions as well as their various classifications.
One of the rapidly developing techniques of image registration is the so-called
hyperspectral imaging, which is used, inter alia, in biology and medicine. Issues
related to the development of profiled software allowing for the hyperspectral
analysis of biological and medical images is the goal of this monograph.

Zygmunt Wróbel

v
Preface

Modern methods of infrared, visible light or UV-light imaging are used in many
fields of science, starting with astronomy through biophysics, physics, geography
and ending with medicine. One such method allowing for imaging in a wide
wavelength spectrum is hyperspectral imaging. The use of this type of imaging
provides ample opportunities not only in terms of the qualitative assessment of
acquired images but also in their quantification. The possibility of quantitative
assessment is the result of analysis performed in the software provided with
hyperspectral cameras. However, due to the large amount of data, this software has
numerous limitations and is user-friendly in a limited way. On the other hand, there
are well-known methods of 2D image analysis and processing. Their implemen-
tation in hyperspectral imaging is not an easy task. Apart from the need to extend
2D images into the third dimension (in which respect there are known methods of
image analysis and processing, but in visible light), there remains the issue of
optimization. It concerns optimization of computational complexity, optimization
of analysis time and performance of preliminary calculations commonly used by
users. The tasks that need to be solved by the users analysing hyperspectral medical
images are also specific by their very nature. The specificity of these images stems
directly from the inter-individual variability in patients and thus the images anal-
ysed. For this reason, for almost any task in question, object segmentation,
comparison, calculation of characteristics, individual profiling of an algorithm are
extremely important. Dedicated algorithms enable to conduct automated, repeatable
measurements of, for example, a specific disease entity. However, profiled
(dedicated) algorithms also have drawbacks—data overfitting. Therefore, these
methods must be tested on images acquired under different conditions, with dif-
ferent hardware settings and for different operators, for example, a hyperspectral
camera. Only in this case, it is certain that the proposed new algorithm will meet the
requirements of universality when it comes to the data source and manner of
acquisition and will be profiled for a particular application. Therefore, the key
element is not only to propose new dedicated methods of hyperspectral image
analysis and processing but also to try to implement them in practice and test their
properties.

vii
viii Preface

The presented methods of analysis and processing of hyperspectral medical

images have been tested in practice in the Matlab® environment. The applied source
code is attached to this monograph. The reader does not need to rewrite its frag-
ments from the text. The source code is also described in detail in the monograph.
The monograph is intended for computer scientists, bioengineers, doctoral stu-
dents and dermatologists interested in contemporary analysis methods. It can also
be used to teach senior students of engineering studies related to computer science
if the price of the book does not constitute a barrier. For the full understanding
of the issues discussed, it has been assumed that the reader knows the basic methods
and matrix operations in Matlab and knows the basic functions of Image
Processing, Signal Processing and Statistics Toolboxes. Finally, other group of
readers who want to know the way to solve the discussed problems in the field of
image analysis and processing in Matlab may become interested in this monograph.

Sosnowiec, Poland Robert Koprowski

Acknowledgments

First, I would like to thank Dr. Sławomir Wilczyński from the Medical University of
Silesia in Katowice for the inspiration and consultations in the area of dermatological
issues covered in the monograph. I would also like to thank Mr Raphael Stachiewicz
from Enformatic Sp. z. o. o., Poland, for providing the hyperspectral camera. It
allowed me to test the developed algorithms and perform measurements for objects
other than medical ones.
I also thank Prof. Zygmunt Wróbel from the University of Silesia in Katowice,
Poland, for his daily support and consultations.

ix
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Purpose and Scope of the Monograph . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4.1 Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4.2 Evaluation of the Classifier Quality. . . . . . . . . . . . . . . . . . . 4
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Image Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Image Pre-Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 Affine Transformations of the Image . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Image Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.1 Non-Adaptive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.2 Adaptive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Image Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4 Preliminary Version of the GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5 Block Diagram of the Discussed Transformations . . . . . . . . . . . . . 35
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1 Diagnostic Expectations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Tracking Changes in the Shape of an Object . . . . . . . . . . . . . . . . . 40
4.3 Basic Analysis of Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Block Diagram of the Discussed Transformations . . . . . . . . . . . . . 67
4.5 Measurement of Additional Features . . . . . . . . . . . . . . . . . . . . . . . 67
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

xi
xii Contents

5 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.1 Decision Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.2 Naive Bayes Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.3 Discriminant Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.4 Support Vector Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.5 Receiver Operating Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.6 Pitfalls and Limitations of Classification . . . . . . . . . . . . . . . . . . . . . 98
5.7 Blok Diagram of the Discussed Transformations . . . . . . . . . . . . . . 105
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6 Sensitivity to Parameter Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.1 Respositioning the ROI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.2 Resizing the ROI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.3 Rotation of the ROI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4 The Other Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Selected Symbols

m,u Row
n,v Column
i Number of the image in a sequence—wavelength
L(m,n,i) Point of the matrix of the image L
L(C) Image resulting from conditional operations
L(D) Image which complements the image L
LBIN Binary image
LGRAY Image in grey levels

Erosion symbol
 Dilation symbol
! Neighbourhood type
h Filter mask
pr Threshold
δg Measurement error
σ Standard deviation of the mean
θ Angle of inclination of the filter
A Amplitude
ACC Accuracy
B Number of bits of an image
FN False negative
FP False positive
I Number of images in a sequence
M Number of rows in a matrix
N Number of columns in a matrix
SE Structural element
SPC Specificity
TN True negative
TP True positive
TPR True positive rate

xiii
Chapter 1
Introduction

1.1 Purpose and Scope of the Monograph

The purpose of this monograph is to present new and known modified methods of
hyperspectral image analysis and processing and profile them in terms of their
usefulness in medical diagnostics and research, as well as develop quantitative
diagnostic tools that can be used in everyday medical practice. The algorithms
proposed in this monograph have the following characteristics:
• they are fully automatic—they do not require operator intervention, if it is
necessary to provide additional parameters of the algorithm operation, they are
selected automatically,
• the results obtained on their basis are fully reproducible,
• their operation was tested on a group of several thousands of hyperspectral
images,
• they were implemented in Matlab,
• they have an open and tested source code attached to this monograph (in the
form of an external link),
• they can be freely extended and modified—owing to the open source code.
The scope of the monograph includes medical images and, in particular, der-
matological ones. However, they are only used to test the discussed methods. The
scope of the monograph is divided into acquisition, image pre-processing, image
processing and their classification presented in the following chapters.

© Springer International Publishing AG 2017 1

R. Koprowski, Processing of Hyperspectral Medical Images,
Studies in Computational Intelligence 682,
DOI 10.1007/978-3-319-50490-2_1
2 1 Introduction

1.2 Material

Most of the images analysed in this monograph had a resolution

M × N × I = 696 × 520 × 128, where M—the number of rows, N—the number of
columns, I—the number of analysed wavelengths. Images of such or similar res-
olution (dependent on individual camera settings) were acquired with different
hyperspectral cameras. The overwhelming part (approximately 75%) of all 200,000
images was registered using the SOC710-VP Hyperspectral Imager with a colour
resolution B = 12 bits and spectral resolution from 400 to 1000 nm. This camera
enables to register 128 bands (I = 128) and is powered by 12 V. The analysed
images were obtained retrospectively and showed the skin of the hand, forearm, and
other areas of the body recorded for dozens of patients. The patients were subject to
exclusion criteria which were undisclosed skin diseases, fever, cardiac arrhythmias,
seizures, inflammation of the skin and pregnancy. The analysed areas were illu-
minated by sunlight or using 40 W halogen lamps of a constant radiation spectrum
ranging from 400 to 1000 nm. All the patients gave informed consent for the study
which was conducted in accordance with the Declaration of Helsinki. No tests,
measurements or experiments were performed on humans as part of this work. This
monograph only deals with the methods of analysis of their images and diagnostic
utility of the obtained results.

1.3 State of the Art

The subject of hyperspectral image analysis and the imaging method itself has been
known for many years. On the day of writing this monograph, the end of 2016, the
PubMed database contained 1922 publications containing the word “hyperspectral”
in the title or description. Slightly different numbers (the number of articles) were
given by the AuthorMapper database, namely 1825 publications, 18,643 authors
from 6105 institutions. A breakdown by countries, institutions, authors, journals
and subjects (the first 5 are listed) is presented in Table 1.1.
As shown above, the leaders in terms of publications on hyperspectral imaging
are the United States, author Chang Chein-I and the area of Computer Science with
1420, 56, 1299 publications respectively. Image Processing and Computer Vision is
a particularly exploited subject, which is extremely important from the point of
view of this monograph. This subject includes such areas as (the number of pub-
lications is given in parentheses): Signal, Image and Video Processing (39); Journal
of Real-Time Image Processing (37); Reference Recognition and Image Analysis
(27); Hyperspectral Imaging (26); Real-Time Progressive Hyperspectral Image
Processing (26); Neural Computing and Applications (25); Advances in Visual
Computing (22) Image Analysis and Recognition (22); Image and Signal
Processing (22); Multiple Classifier Systems (20); Machine Vision and
Applications (19); Hyperspectral Data Compression (17); Advanced Concepts for
1.3 State of the Art 3

Table 1.1 The first 5 countries, 5 institutions, 5 authors, 5 journals, 5 subjects related to the word
“hyperspectral”
Country United China Germany India France
States
Number of 1420 [1–5] 995 [6–10] 455 [11–15] 347 [16–20] 311 [21–25]
publications
Institution Chinese Zhejiang University of University of Wuhan
Academy of University California Maryland University
Sciences
Number of 162 [26–30] 67 [31–34] 64 [35–40] 47 [41–44] 43 [44–50]
publications
Author Chang, Graña, Manuel Sun, Da-Wen Goodacre, Wang, Liguo
Chein-I [51, [53, 54] [55–58] Royston [59, [61, 62]
52] 60]
Number of 56 29 21 18 18
publications
Journal Precision Journal of the Environmental Analytical Environmental
Agriculture Indian Society of Monitoring and and Earth Sciences
Remote Sensing Assessment Bioanalytical
Chemistry
Number of 144 [63, 64] 140 [65, 66] 100 [67, 68 91 [69, 70] 77 [71, 72]
publications
Subject Computer Life Sciences Artificial Earth Image
Science Intelligence Sciences Processing and
(incl. Robotics) Computer
Vision
Number of 1299 [73, 960 [75, 76] 908 [77, 78] 859 [79, 80] 830 [81, 82]
publications 74]

Intelligent Vision Systems (16); Journal of Signal Processing Systems (16);

Mathematical Morphology and Its Applications to Signal and Image Processing
(15); Remote Sensing Digital Image Analysis (15); Image Analysis (14);
Hyperspectral Image Fusion (13); Hyperspectral Image Processing (12); Journal of
Mathematical Imaging and Vision (11).
When reviewing publications [83–91] in terms of the described research prob-
lems and their solutions, several open issues in the field of hyperspectral image
analysis can be observed:
• the need for profiling methods of image analysis and processing to a particular
research problem,
• lack of universal methods of analysis and
• lack of or limited availability of source codes.
Therefore, this monograph describes a sample application for the analysis and
processing of hyperspectral images. The application was profiled to the area of
biomedical engineering, and includes both known and new algorithms for image
analysis and processing.
4 1 Introduction

The discussed scope of biomedical engineering involves the use of hyperspectral

cameras in dermatology. These issues have been partly presented in several pub-
lications [92–96]. Some of them are not profiled to solve a particular segmentation
issue and do not address the issue of the algorithm sensitivity to parameter changes
or the impact of different methods of acquisition on the results obtained.
Accordingly, the analysis of the impact of acquisition on the results obtained, at the
full automation of the proposed algorithm, constitutes another area (chapter) of this
monograph.

1.4 Basic Definitions

Basic definitions apply to two issues:

• orientation of coordinate systems and
• assessment of the classifier quality.
They are described in the following sections.

1.4.1 Coordinate System

The orientation of the coordinate system is strongly dependent on the individual

settings of the camera relative to the object, frame of reference. Regardless of the
individual camera setting, to which all the described algorithms cannot be sensitive,
it was assumed that the size of each image sequence would be defined by the
number of rows M numbered from one, the number of columns N and the number
of individual wavelengths I. The numbering from one and not zero, as in the case of
well-known programming languages C++, C#, results from the adopted nomen-
clature and numbering in Matlab, Scilab or Octave. As a result, it was adopted in
this monograph—Fig. 1.1 and Fig. 1.2.
The presented coordinate system (Fig. 1.1) will be used for all the presented
analyses and algorithms. When an image (matrix) is a single 2D matrix, dimension
I will be 1.

1.4.2 Evaluation of the Classifier Quality

Classifiers were usually induced by using the training data representing 2/3 of the
total number of data. The remaining 1/3 of the data was used to test the classifier
quality [96]. The training and test data were divided randomly. In the cases pre-
sented in this monograph, the division can be distorted. This is due to the fact that
1.4 Basic Definitions 5

Fig. 1.1 The following

symbols were adopted in the

coordinate system: M number
of rows, N number of
columns, I number of
wavelengths (random colours
of individual pixels were
adopted) 


 


 


 





Fig. 1.2 Orientation of the

adopted coordinate system for

a colour image—RGB (pixel
colours correspond to R, G
and B components)

1-R 


2-G 
 

3-B 

 

both the training and test vectors result from calculations for a sequence of images.
The number of images in the sequence to be analysed is affected by the operator.
Therefore, it does not have to be a strict division into 1/3 and 2/3. In each case,
evaluation of the classifier quality was based on determination of four values:
• False Positive (FP)—cases incorrectly classified as positive,
• False Negative (FN)—cases incorrectly classified as negative,
• True Positive (TP)—positive cases classified correctly,
• True Negative (FN)—negative cases classified correctly.
On their basis, sensitivity TPR and specificity SPC were defined as [97, 98]:
6 1 Introduction

TP
TPR ¼  100% ð1:1Þ
TP þ FN
TN
SPC ¼  100% ð1:2Þ
TN þ FP

and accuracy ACC:

TP þ TN
ACC ¼  100% ð1:3Þ
TP þ TN þ FP þ FN

The parameters SPC and TPR will be the basis for creating the receiver operating
characteristic curves (ROC), which are graphs of changes in TPR as a function of
100-SPC [99]. Additionally, the area under the curve (AUC) will be calculated
[100].

References

1. Wang, Liguo; Zhao, Chunhui, Introduction of Hyperspectral Remote Sensing Applications,

Hyperspectral Image Processing: 283–308, (2016).
2. Nguyen, Hien Van; Banerjee, Amit; Burlina, Philippe; Broadwater, Joshua; Chellappa,
Rama, Tracking and Identification via Object Reflectance Using a Hyperspectral Video
Camera, Machine Vision Beyond Visible Spectrum (2011-01-01) 1: 201–219, January 01,
2011.
3. Miguel, Agnieszka; Riskin, Eve; Ladner, Richard; Barney, Dane, Near-lossless and lossy
compression of imaging spectrometer data: comparison of information extraction perfor-
mance, Signal, Image and Video Processing (2012-11-01) 6: 597–611, November 01, 2012.
4. Le, Justin H.; Yazdanpanah, Ali Pour; Regentova, Emma E.; Muthukumar, Venkatesan A
Deep Belief Network for Classifying Remotely-Sensed Hyperspectral Data, Advances in
Visual Computing (2015-01-01): 9474, January 01, 2015.
5. Alam, Mohammad S.; Sakla, Adel Automatic Target Recognition in Multispectral and
Hyperspectral Imagery Via Joint Transform Correlation, Wide Area Surveillance
(2012-09-11) 6: 179–206, September 11, 2012.
6. Miller, Corey A.; Walls, Thomas J., Hyperspectral Scene Analysis via Structure from
Motion, Advances in Visual Computing (2015-01-01): 9474, January 01, 2015.
7. Zou, Xiaobo; Zhao, Jiewen, Hyperspectral Imaging Detection, Nondestructive Measurement
in Food and Agro-products (2015-01-01): 127–193, January 01, 2015.
8. Chen, Yu-Nan; Sun, Da-Wen; Cheng, Jun-Hu; Gao, Wen-Hong, Recent Advances for Rapid
Identification of Chemical Information of Muscle Foods by Hyperspectral Imaging Analysis,
Food Engineering Reviews (2016-04-04): 1–15, April 04, 2016.
9. Wang, Liguo; Zhao, Chunhui, Introduction of Hyperspectral Remote Sensing Applications,
Hyperspectral Image Processing (2016-01-01): 283–308, January 01, 2016.
10. Zhang, Baohua; Fan, Shuxiang; Li, Jiangbo; Huang, Wenqian; Zhao, Chunjiang; Qian, Man;
Zheng, Ling, Detection of Early Rottenness on Apples by Using Hyperspectral Imaging
Combined with Spectral Analysis and Image Processing, Food Analytical Methods
(2015-09-01) 8: 2075–2086, September 01, 2015.
References 7

11. Behmann, Jan; Mahlein, Anne-Katrin; Paulus, Stefan; Dupuis, Jan; Kuhlmann, Heiner;
Oerke, Erich-Christian; Plümer, Lutz, Generation and application of hyperspectral 3D plant
models: methods and challenges, Machine Vision and Applications (2015-10-03): 1–14,
October 03, 2015.
12. Lausch, Angela; Pause, Marion; Merbach, Ines; Zacharias, Steffen; Doktor, Daniel; Volk,
Martin; Seppelt, Ralf, A new multiscale approach for monitoring vegetation using remote
sensing-based indicators in laboratory, field, and landscape, Environmental Monitoring and
Assessment (2013-02-01) 185: 1215–1235, February 01, 2013.
13. Martin, Ron; Thies, Boris; Gerstner, Andreas OH, Hyperspectral hybrid method classifi-
cation for detecting altered mucosa of the human larynx, International Journal of Health
Geographics (2012-06-21) 11: 1–9, June 21, 2012.
14. Bigdeli, Behnaz; Samadzadegan, Farhad; Reinartz, Peter A Multiple SVM System for
Classification of Hyperspectral Remote Sensing Data, Journal of the Indian Society of
Remote Sensing (2013-12-01) 41: 763–776, December 01, 2013.
15. Wang, Ke; Gu, XingFa; Yu, Tao; Meng, QingYan; Zhao, LiMin; Feng, Li Classification of
hyperspectral remote sensing images using frequency spectrum similarity, Science China
Technological Sciences (2013-04-01) 56: 980–988, April 01, 2013.
16. Mookambiga, A.; Gomathi, V.Comprehensive review on fusion techniques for spatial
information enhancement in hyperspectral imagery, Multidimensional Systems and Signal
Processing (2016-04-27): 1–27, April 27, 2016.
17. Chutia, Dibyajyoti; Bhattacharyya, Dhruba Kumar; Kalita, Ranjan; Goswami, Jonali; Singh,
Puyam S.; Sudhakar, S., A model on achieving higher performance in the classification of
hyperspectral satellite data: a case study on Hyperion data, Applied Geomatics (2014-09-01)
6: 181–195, September 01, 2014.
18. Prabhu, N.; Arora, Manoj K.; Balasubramanian, R.Wavelet Based Feature Extraction
Techniques of Hyperspectral Data, Journal of the Indian Society of Remote Sensing
(2016-01-29): 1–12, January 29, 2016.
19. Prabhakar, M.; Prasad, Y. G.; Rao, Mahesh N., Remote Sensing of Biotic Stress in Crop
Plants and Its Applications for Pest Management, Crop Stress and its Management:
Perspectives and Strategies (2012-01-01): 517–545, January 01, 2012.
20. Vetrekar, N. T.; Gad, R. S.; Fernandes, I.; Parab, J. S.; Desai, A. R.; Pawar, J. D.; Naik, G.
M.; Umapathy, S., Non-invasive hyperspectral imaging approach for fruit quality control
application and classification: case study of apple, chikoo, guava fruits, Journal of Food
Science and Technology (2015-11-01) 52: 6978–6989, November 01, 2015.
21. Licciardi, Giorgio Antonino Hyperspectral Data in Urban Areas Encyclopedia of Earthquake
Engineering (2015-01-01): 1155–1164, January 01, 2015.
22. Luo, Bin; Chanussot, Jocelyn Supervised Hyperspectral Image Classification Based on
Spectral Unmixing and Geometrical Features, Journal of Signal Processing Systems
(2011-12-01) 65: 457–468, December 01, 2011.
23. Xia, Junshi; Chanussot, Jocelyn; Du, Peijun; He, Xiyan, Rotation-Based Ensemble
Classifiers for High-Dimensional Data, Fusion in Computer Vision (2014-03-26): 135–
160, March 26, 2014.
24. Hernández-Sánchez, Natalia; Moreda, Guillermo P.; Herre-ro-Langreo, Ana;
Melado-Herreros, Ángela, Assessment of Internal and External Quality of Fruits and
Vegetables, Imaging Technologies and Data Processing for Food Engineers (2016-01-01):
269–309, January 01, 2016.
25. Chen, Weishi; Guillaume, Mireille, HALS-based NMF with flexible constraints for
hyperspectral unmixing, EURASIP Journal on Advances in Signal Processing
(2012-03-05) 2012: 1–14, March 05, 2012.
26. Zhang, Bing; Yang, Wei; Gao, Lianru; Chen, Dongmei Real-time target detection in
hyperspectral images based on spatial-spectral information extraction EURASIP Journal on
Advances in Signal Processing (2012-07-13) 2012: 1–15, July 13, 2012.