Texture analysis using complex system models:

fractal dimension, swarm systems and

non-linear diffusion
Bruno Brandoli Machado, Jose Fernando Rodrigues Junior
University of São Paulo (ICMC/USP)
13566-590, São Carlos, SP, Brasil
emails: brunobrandoli@gmail.com, junio@icmc.usp.br

Abstract—Texture is one of the primary visual attributes used of the images [9]. The idea comes from concepts on mathe-
to describe patterns found in nature. Several texture analysis matical morphology, which describes an image by evolving
methods have been used as powerful tools for real applications morphological operations with different sizes of structuring
involving computer vision. However, existing methods do not
successfully discriminate the complexity of texture patterns. Such elements [10], a useful technique to handle shapes in textures.
methods disregard the possibility of describing image structures Statistical methods represent textures by the spatial dis-
by fractal dimension. Fractality-based measures allow a non- tribution of the gray-level pixels in the image. One of the
integer geometric interpretation with applications in areas such best methods of this category, and still very popular, is co-
as mathematics, physics, and biology. The central hypothesis of occurrence matrices [11], [12]. In the same line, Dmitry
this work is that textures can be described as irregular fractal
surfaces due to their complex geometry. Pushing the limits of Chetverikov [13] introduced the technique named interaction
the state-of-the-art, the results achieved in the four methodologies map. Similarly, Ojala et. al. [14] proposed a method that
described in this work demonstrated the potential of using texture describes images based on the occurrence of gray values on
features in tasks of pattern recognition. The contributions of this circular local neighborhoods; it is named local binary patterns
work shall support significant advances in materials engineering, (LBP). Xiaoyang et al. [15] extended the idea of LBP to local
computer vision, and agriculture.
ternary pattern (LTP), which considers the magnitude of pixel
Keywords-texture analysis; fractal dimension; swarm system; derivatives along with its sign to generate the ternary code. In
non-linear diffusion; complex networks the work of Hadid et al. [16], the authors present a comparative
study using 13 variants of local binary patterns for gender
I. I NTRODUCTION classification.
1 Model-based methods were proposed with the assumption
Texture is an important visual attribute in computer vision
that textures can be represented by mathematical models,
with many areas of applications. Recently, texture analysis
including stochastic models of Markov random fields [17],
has been widely applied to remote sensing [1], [2], industrial
[18] and fractality [19]. Particularly, fractal geometry has
inspection [3], [4], medical image analysis [5], [6], face recog-
drawn great attention on the task of describing textures [20].
nition [7], [8], among many others. Although the human visual
The main reason is because fractal geometry is able to describe
system can easily discriminate textural patterns, the description
irregular or fragmented shapes of natural phenomena, as well
by automatic methods has been a great challenge. Indeed, there
as other complex objects that traditional Euclidean geometry
is no universally accepted definition of texture. It is usually
is not able to describe. The fractal concept is expressed by the
referred to as a repetitive pattern that can vary according to
time-domain statistical scaling laws and is characterized by
the size, which produces different tactile sensations associated
the power-law behavior of physical systems. This idea makes
with roughness, coarseness, and regularity. Furthermore, tex-
use of the geometrical interpretation of objects and takes into
ture patterns are related to the physical properties of surfaces
account the degree of regularity of the structure related to
present in images, making them a powerful tool for image
its physical behavior [21]. Recently, Goncalves et al. [22]
analysis.
proposed a hybrid fractal-swarm method for texture analysis.
Texture analysis has been an active research field in the last
Most of the methods used in texture analysis are restricted to
decade. The proposed methods have been grouped according
the analysis of the spatial relations over small neighborhoods,
to the mathematical aspects used to handle the patterns present
or they are based on the extraction of global features of
in the images. There are five major categories: structural, sta-
the whole image on a single scale. As a consequence, they
tistical, spectral, model-based, and agent-based. The structural
do not perform well in high-complex geometry textures. In
methods rely on primitives that provide a symbolic description
contrast, spectral methods or signal processing methods, in-
1 This manuscript is related to a Ph.D. thesis. The thesis can be downloaded
cluding Fourier analysis [23], Gabor filters [24], [25], wavelet
at http://www.teses.usp.br, by clicking on the option Advanced Search; select transform methods [26], [27], were developed inspired by
Doctoral Thesis, and then type the first two names of the main author. evidence that the human visual system describes images by
the frequency domain. However, the Fourier transform lacks to the method proposed in the first contribution, each agent
spatial information, what impairs its potential for image de- is able to move to the higher intensities, as well as to lower
scription. Although Gabor filters present joint image resolution ones. Although we can find the minima and maxima of images
in both the spatial and frequency domains, they do not describe directly, the underlying idea is to characterize the path of
well coarse textures; this is because the energy of such movement during the evolution process. Our method differs
textures is concentrated in subimages of lowest frequencies from the original artificial crawlers since we quantify the
[28]. In contrast with Gabor filters, wavelets decompose a state of the swarm system after stabilization by employing
texture image into a set of frequency channels. However, the Bouligand-Minkowski fractal dimension method [20]. In
wavelet analysis is basically a linear analysis and suffers from the method, the energy information was considered the most
uniformly-poor resolutions over different scales and from its important attribute due to its capacity of representing the inter-
non-data adaptive nature, since the same wavelet basis is used action between the movement of agents and the environment.
to analyze all the data [29]. This work was published in the journal Physica A: Statistical
Mechanics and its Applications [22] and, an example of the
II. G OALS AND C ONTRIBUTIONS path configuration for 1,000 artificial crawlers is presented in
The goal of this work is to propose solutions to overcome Figure 1.
the drawbacks of current texture analysis methods, as observed
in the literature. To reach this goal, we have proposed new
methods for texture description based on complex systems,
including fractal dimension, swarm systems, non-linear dif-
fusion models, and complex networks. We have addressed
two main issues found in typical statistical and model-based
texture descriptors: (1) the lack of a multiscale representation
to capture the richness of local features in different levels
of observation; and, (2) although fractal descriptors proved
to be a promising texture descriptor, current methods do not (a)
explicitly consider the neighborhood relation in terms of the
gray levels of the texture. In this work, we evaluated our
methods over four texture benchmarks: Brodatz [30], Vistex
[31], Outex [32] and Usptex [33]. In addition, our methods
were compared with state-of-the-art texture methods. The
motivation to develop our methods is that they are suitable for
real-world applications, as in material engineering and agri-
culture, leading to better or automatic decisions. As we show
along the text, we focus on nanomaterial quality assessment
and on plant disease identification in soybean leaves.
The first contribution of this work is a methodology for (b) r = 2 (c) r = 3
texture description. We carefully investigate all the steps of
Fig. 1. In (a) artificial crawlers are mapped onto a 3D space by converting
a swarm system method named artificial crawler, proposed the final position and the energy into a point in the surface. (b) e (c) show an
first by [34], [35]. The original method was proposed to illustration of the dilation process for the fractal dimension.
have the iterative crawling step moving only in the direction
of maximum pixel intensity, thus characterizing regions of The third contribution of this work is a multiscale texture
high intensities in the image. However, in texture analysis, descriptor based on non-linear diffusion. Many recent texture-
regions of low intensities are as important as regions of analysis methods are developed to extract measures on a single
high intensities. For this reason, we proposed a new rule scale. In contrast, we assume that an image texture reveals
of movement that also moves the agents in the direction different structures according to the scale of observation, so
of lower intensity. We developed an improved method for that the scale concept of multiscale representation is of crucial
assessing the quality of the silk fibroin scaffolds [36] with importance [38]. Thus, we iteratively represent the original
two rules of movement: maxima and minima. Our goal is to image in a set of new images. Inspired by the work of
provide an effective method to support visual analysis, thus Yves Meyer [39], where images are combinations of both
reducing the subjectivity of the human analysis. We evaluated oscillatory (texture) and geometrical (cartoon) patterns, we get
the potential of the silk fibroin by including glycerol in the two components for the derivative images. At each iteration
solution during scaffold formation. This work was published step, we estimate the average and deviation of the Bouligand-
in the Computational Science and Discovery Journal [37]. Minkowski descriptors computed over the two components.
The second contribution of this work is an extended method- We then combine the measures from both components to
ology based on the energy information of the artificial crawler compose our feature vector. The Bouligand-Minkowski fractal
swarm system extracted from two rules of movement. Similar dimension is adopted here due to its precision in quantifying
structural properties. Experimental results over four well- 3.
known texture datasets reveal a superior performance of our
method. This work was first published in the Proceedings of
the 13th International Conference on Advanced Concepts for
Intelligent Vision Systems with Gabor filters [40]. Later, a new
research involving fractal descriptors was submitted to the
Pattern Recognition Letters Journal [38] and the main ideia
of the proposal is presented in Figure 2.

Fig. 3. Screenshots of the BioLeaf application, and it can be down-

loaded freely from GooglePlay at https://play.google.com/store/apps/details?
id=upvision.bioleaf.

Finally, the fifth contribution of this work is a com-

plex network approach for particle agglomeration analysis
in nanoscale images. In this work, instead of assuming that
a nanoscale image is a textured surface, we have modeled
the nanoparticles like vertices of a graph, while connections
are created according to a thresholding for density estimation
Fig. 2. Schematic representation of our proposal. The dotted-line frames
over a certain radius. For each nanoparticle, we calculate its
correspond to the images obtained by anisotropic diffusion over the scale density. Two particles are linked, defining and edge, only if
space, and t stands for the different scales. The image decomposition is their distance is smaller than a given radius and its density is
applied to each image of the dataset resulting in two component images:
cartoon and texture. Then, the fractal dimension is estimated from each image,
higher than a given threshold. This work has been submitted
and a feature vector is concatenated with the mean and standard deviation of to the Information Sciences Journal [43]. Furthermore, this
the fractal dimension. work was integrated into an expert system (see Figure 4),
named NanoImage Analyzer, and it was submitted for soft-
The fourth contribution of this work is the application of ware registration also at INPI. NanoImage Analyzer can be
the multiscale proposal shown in Figure 2. In this case, we downloaded freely at http://nanoanalyzer.icmc.usp.br.
assume that a leaf image presents different details according
to the scale of analysis, which is suitable to describe fractal- III. C ONCLUSION
like structures as observed in leaves. Accordingly, we propose Texture description has always been a challenging task in
a multiscale fractal descriptor that is applied over derivative image analysis and computer vision. Texture analysis methods
images obtained by means of anisotropic diffusion. In our have emerged as important tools for real-world applications,
methodology, the gray levels of an image correspond to the this is because they can learn patterns from objects or regions
energy diffusion at different levels represented in new deriva- with annotated examples. Typically, such methods aim to
tive images. We split each new derivative into geometrical and measure raw image pixels into a discriminant data space.
oscillatory parts. Subsequently, we estimate the Bouligand- Despite the achievements, effective feature description is still
Minkowski fractal dimension of each component and combine a challenge. In this work, we introduced novel methods for
the features to perform texture classification. Experiments in- texture description applied over domains ranging from agri-
dicate that our approach can successfully identify soybean leaf culture to nanomaterials. We have achieved promising results,
diseases and can also be used as a front-end application to non- as previously discussed, with the aim of demonstrating that
experts or agronomists. In addition, our method was compared texture, as found in natural settings (leaves and nanoparticles),
to other traditional texture methods showing superiority for is of great potential in image analysis and computer vision.
recognition of soybean leaf diseases. This work [41] has been In conclusion, this work departed from the hypothesis that
integrated to the mobile application called BioLeaf - Foliar the use of texture information can improve systems that
Analysis [42], which was submitted for software registration depend on image analysis. We tested this initial assumption
at the Brazilian agency National Institute of Industrial Property over several domains using different proposed methods. Our
(INPI). Both the methodology and the mobile application results, as discussed in the previous section, demonstrated
description were submitted to the Computer Electronics and that, indeed, multiscale image representation has potential in
Agriculture Journal. Details of the tool are presented in Figure enhancing the discriminatory power when the scale is taken
 (a) NanoImage Analyzer can be downloaded at http://nanoanalyzer.icmc.usp.br

(b) (c)
Fig. 4. (a) Tool NanoImageAnalyzer implemented for analysis of nanoparticles agglomeration in nanoscale images. In second row, we show the nanoparticle
image modeled as a complex network proposed in this work. (b) Density of each nanoparticle, represented by the heatmap colors, and connections of the
resulting complex network. (c) Zoomed-in regions as indicated in (b).

into account. Future directions of the feature design are lead by Physics. Conference Series (Online), v. 410, p. 012066,
advances in deep neural network construction, that has brought 2013.
a significant breakthrough for feature description. • GONCALVES, W.N., MACHADO, B.B., BRUNO, O.M.
Texture descriptor combining fractal dimension and arti-
IV. S OFTWARE R EGISTRATION AND P UBLICATIONS ficial crawlers. Physica A, v. 395, p. 358-370, 2014.
• MACHADO, B.B., GONCALVES, W.N., BRUNO, O.M.
The work of my Ph.D. was published in the following Artificial crawler model for texture analysis on silk fi-
refereed publications, and it was submitted, in the form of broin scaffolds. Computational Science & Discovery, v.
software registration, to the National Institute of Industrial 7, p. 015004, 2014.
Property (INPI). • GONCALVES, W.N., MACHADO, B.B., BRUNO, O.M.
Dynamic texture recognition based on complex networks.
Journal articles Neurocomputing. v. 153, p. 211-220, 2015.
• MACHADO, B.B., CASANOVA, D., GONCALVES, W. • MACHADO, B.B., GONCALVES, W.N., ARRUDA,
N., BRUNO, O. M. Partial differential equations and M.S., RODRIGUES, J.F.J. Multiscale fractal descriptors
fractal analysis to plant leaf identification. Journal of using anisotropic diffusion of Perona-Malik for texture
 analysis. Pattern Recognition Letter. (submitted) leaf-texture analysis. In: VII Workshop de Visão Com-
• SCABINI, L.F.S., FISTAROL, D., CANTERO, putacional. Curitiba, Paraná. Anais do VII Workshop de
S.V.A.B.C, RODRIGUES, J.F.J., MACHADO, B.B., Visão Computacional. Curitiba : Omnipax, 2011. p. 155-
GONCALVES, W.N. Angular Measures of Complex 160.
Networks for Boundary Shape Analysis, Expert Systems • MACHADO, B.B., GONCALVES, W.N., BRUNO, O.M.
With Applications. (submitted) Enhancing the texture attribute with partial differential
• MACHADO, B.B., ORUE, J., ARRUDA, M.S., SAN- equations: a case of study with Ga-bor filters. In: ACIVS
TOS, C.V., SARATH, D.S., SILVA, G.G., GONCALVES, - Advanced Concepts for Intelligent Vision Systems.
W.N., PISTORI, H., ROEL, A.R., RODRIGUES, J.F.J. Ghent, Belgium. Lecture Notes in Computer Science.
BioLeaf: a professional mobile application to measure Berlin: Springer, 2011. v.6915. p. 337-348.
foliar damage caused by insect herbivory, Computer • MACHADO, B.B., CASANOVA. D, GONCALVES,
Electronics and Agriculture. (submitted) W.N., BRUNO, O.M. Partial differential equations and
• MACHADO, B.B., ORUE, J., ARRUDA, M.S., SAN- fractal analysis to plant leaf identification. In: Interna-
TOS, C.V., SARATH, D.S., SILVA, G.G., GONCALVES, tional Conference on Mathematical Modeling in Physical
W.N., PISTORI, H., ROEL, A.R., RODRIGUES, Sciences. Budapest, Hungary, 2012, p. 207-207.
J.F.J. Quantificação automática da área foliar usando • MACHADO, B.B., GONCALVES, W.N., BRUNO, O.M.
reconstrução por curvas de Bézier, Revista Pesquisa Material quality assessment of silk fibroin nanofibers
Agropecuária Brasileira. (submitted) based on swarm intelligence. In: International Confer-
• MACHADO, B.B., ORUE, J., ARRUDA, M.S., ence on Mathematical Modeling in Physical Sciences.
GONCALVES, W.N., MOREIRA, R., RODRIGUES, Budapest, Hungary, 2012. p. 241-241.
J.F.J. A Complex Network Approach for Nanoparticle • GONCALVES, W. N., MACHADO, B.B., BRUNO, O.
Agglomeration Analysis in Nanoscale Images, Computers M. Dynamic texture recognition based on complex net-
and Chemical Engineering. (submitted) works. In: International Conference on Mathematical
• PIRES, R.L.D., GONCALVES, D.N., ORUE, J., Modeling in Physical Sciences, 2012, Bristol. Interna-
KANASHIRO, W., RODRIGUES, J.F.J., MACHADO, tional Conference on Mathematical Modeling in Physical
B.B., GONCALVES, W.N. Local descriptors for soybean Sciences. Budapest, 2012. p. 202-202.
disease recognition, Computer and Electronics in Agri- • SARATH, D. S., SILVA, G.G., ROEL, A. R., PERUCA,
culture, v. 125, p. 48-55, 2016. R. D., MACHADO, B.B., PISTORI, H. Quantificação
• MACHADO, B.B., GONCALVES, W.N., ARRUDA, automática da área foliar na cultura da soja usando
M.S., WELBER, B., RODRIGUES, J.F.J. Identification segmentação de imagens coloridas. ISBN 978-85-69929-
of soybean leaf diseases using multiscale fractal descrip- 00-0, In: X Congresso Brasileiro de Agroinformática,
tors, Computer and Electronics in Agriculture. (submit- 2015, Ponta Grossa - PR, 2015. p. 102-108.
ted) • ARRUDA, M.S., MACHADO, B.B., GONCALVES,
• RODRIGUES, J.F.J., ZAINA, A.M.L., OLIVEIRA, M.C., W.N., DIAS, J.H.P., CULLEN, L., GARCIA, C.C., RO-
MACHADO, B.B., TRAINA, A. A survey on information DRIGUES, J F.J. Thermal Image Segmentation in Studies
visualization in light of vision and cognitive sciences: rec- of Wildlife Animals. In: Workshop de Visão Computa-
ommendations for effective design, The Visual Computer. cional, 2015, São Carlos. XI Workshop de Visão Com-
(submitted) putacional, 2015. p. 204-209.
• GONCALVES, D. N., SILVA, L. A., ARAUJO, R. F. S.,
Software registration MACHADO, B.B., GONCALVES, W. N. Texture anal-
ysis using local fractal dimension of complex networks.
• MACHADO, B.B., ORUE, J., ARRUDA, M.S., SAN-
In: Workshop de Visão Computacional, 2015, São Carlos.
TOS, C.V., SARATH, D.S., SILVA, G.G., GONCALVES,
XI Workshop de Visão Computacional, 2015, p. 236-241.
W.N., PISTORI, H., ROEL, A.R., RODRIGUES, J.F.J.
• PIRES, R. D. L., KANASHIRO, W. E. S., GONCALVES,
BioLeaf - Foliar Analysis, 2016. (deposited)
W. N., MACHADO, B.B., ARRUDA, M. S., ORUE, J.
• MACHADO, B.B., ARRUDA, M.S., ORUE, J., SAN-
P. M. Identification of foliar soybean diseases using local
TOS, C.V., GONCALVES, W.N., RODRIGUES, J.F.J.
descriptors. In: Workshop de Visão Computacional, 2015,
DropLeaf - Deposition Analysis, 2016. (deposited)
São Carlos. XI Workshop de Visão Computacional, 2015,
• MACHADO, B.B., SCABINI, L.F.S., ARRUDA, M.S.,
p. 242-247.
ORUE, J., GONCALVES, D.N., GONCALVES, W.N.,
• GONCALVES, D. N., SILVA, N., ARRUDA, M.S.,
MOREIRA, R., RODRIGUES, J.F.J. NanoImage Ana-
MACHADO, B.B., GONCALVES, W. N. Animal species
lyzer, 2016. (deposited)
recognition using deep learning. 2016. (submitted)
Conference articles ACKNOWLEDGMENT
• MACHADO, B.B., GONCALVES, W.N., BRUNO, O.M. The authors are grateful to the funding agencies by the fi-
Image decomposition via anisotropic diffusion applied to nancial support from FAPESP and CNPq, under grant numbers
2011/02918-0 and 142098/2011-6, respectively. [22] W. N. Goncalves, B. B. Machado, and O. M. Bruno, “Texture descriptor
combining fractal dimension and artificial crawlers,” Physica A: Statis-
R EFERENCES tical Mechanics and its Applications, vol. 395, pp. 358–370, 2014.
[23] R. Azencott, J.-P. Wang, and L. Younes, “Texture classification using
[1] T. Corpetti and O. Planchon, “Front detection on satellite images windowed fourier filters,” IEEE Trans. Pattern Anal. Mach. Intell.,
based on wavelet and evidence theory: Application to the sea breeze vol. 19, pp. 148–153, February 1997.
fronts,” Remote Sensing of Environment, vol. 115, no. 2, pp. 306–324, [24] D. Gabor, “Theory of communication,” Journal of Institute of Electronic
2011. [Online]. Available: http://www.sciencedirect.com/science/article/ Engineering, vol. 93, pp. 429–457, November 1946.
pii/S003442571000266X [25] F. Bianconi and A. Fernández, “Evaluation of the effects of gabor
[2] M. Gong, Y. Li, L. Jiao, M. Jia, and L. Su, “Sar change detection based filter parameters on texture classification,” Pattern Recognition, vol. 40,
on intensity and texture changes,” Journal of Photogrammetry and no. 12, pp. 3325–3335, 2007.
Remote Sensing, vol. 93, no. 0, pp. 123–135, 2014. [Online]. Available: [26] I. Daubechies, Ten lectures on wavelets. Philadelphia, PA, USA: Society
http://www.sciencedirect.com/science/article/pii/S0924271614001051 for Industrial and Applied Mathematics, 1992.
[3] D. Kim, J. J. Liu, and C. Han, “Determination of steel quality based [27] S. Mallat and S. Zhong, “Characterization of signals from multiscale
on discriminating textural feature selection,” Chemical Engineering edges,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 14, no. 7, pp.
Science, vol. 66, no. 23, pp. 6264–6271, 2011. [Online]. Available: 710–732, 1992.
http://www.sciencedirect.com/science/article/pii/S0009250911006373 [28] Q. Xu, H. Wu, and Y. Q. Chen, “Statistical multiscale blob features
[4] C. S. Tsang, H. Y. Ngan, and G. K. Pang, “Fabric inspection based for classifying and retrieving image texture from large-scale databases,”
on the elo rating method,” Pattern Recognition, vol. 51, no. 3, Journal of Electronic Imaging, vol. 19, no. 4, pp. 043 006–043 006–7,
pp. 378–394, 2016. [Online]. Available: http://www.sciencedirect.com/ 2010.
science/article/pii/S0031320315003532 [29] N. E. Huang, S. R. Long, and Z. Shen, “The mechanism for
[5] C. Serrano and B. Acha, “Pattern analysis of dermoscopic images frequency downshift in nonlinear wave evolution,” ser. Advances
based on markov random fields,” Pattern Recognition, vol. 42, no. 6, in Applied Mechanics, J. W. Hutchinson and T. Y. Wu, Eds.
pp. 1052–1057, 2009. [Online]. Available: http://www.sciencedirect. Elsevier, 1996, vol. 32, pp. 59–117C. [Online]. Available: http:
com/science/article/pii/S003132030800280X //www.sciencedirect.com/science/article/pii/S0065215608700760
[6] N. Zaglam, P. Jouvet, O. Flechelles, G. Emeriaud, and F. Cheriet, [30] P. Brodatz, Textures: A Photographic Album for Artists and Designers.
“Computer-aided diagnosis system for the acute respiratory distress New York: Dover Publications, 1966.
syndrome from chest radiographs,” Computers in Biology and [31] S. Singh and M. Sharma, “Texture analysis experiments with meastex
Medicine, vol. 52, no. 0, pp. 41–48, 2014. [Online]. Available: and vistex benchmarks,” in Proceedings of the Second International
http://www.sciencedirect.com/science/article/pii/S0010482514001450 Conference on Advances in Pattern Recognition, ser. ICAPR ’01. Lon-
[7] Z. Fu, F. Ni, Q. Cao, and Y. Zhao, “The facial texture analysis for don, UK: Springer-Verlag, 2001, pp. 417–424.
the automatic portrait drawing,” Pattern Recognition, vol. 43, no. 3, [32] T. Ojala, T. Mäenpää, M. Pietikäinen, J. Viertola, J. Kyllönen, and
pp. 962–971, 2010. [Online]. Available: http://www.sciencedirect.com/ S. Huovinen, “Outex - new framework for empirical evaluation of texture
science/article/pii/S0031320309002647 analysis algorithms,” in Proc. 16th International Conference on Pattern
[8] R. Mehta, J. Yuan, and K. Egiazarian, “Face recognition using Recognition, 2002, pp. 701–706.
scale-adaptive directional and textural features,” Pattern Recognition, [33] A. R. Backes, D. Casanova, and O. M. Bruno, “Color texture analysis
vol. 47, no. 5, pp. 1846–1858, 2014. [Online]. Available: http: based on fractal descriptors,” Pattern Recognition, vol. 45, no. 5, pp.
//www.sciencedirect.com/science/article/pii/S0031320313004998 1984–1992, 2012. [Online]. Available: http://www.sciencedirect.com/
[9] Y. Chen and E. Dougherty, “Gray-scale morphological granulometric science/article/pii/S0031320311004614
texture classification,” Optical Engineering, vol. 33, no. 8, pp. 2713– [34] D. Zhang and Y. Q. Chen, “Classifying image texture with artificial
2722, 1994. crawlers,” in Proceedings of the IEEE/WIC/ACM International Confer-
[10] J. Serra, Image Analysis and Mathematical Morphology. Orlando, FL, ence on Intelligent Agent Technology, ser. IAT ’04. Washington, DC,
USA: Academic Press, Inc., 1983. USA: IEEE Computer Society, 2004, pp. 446–449.
[11] R. M. Haralick, K. Shanmugam, and I. Dinstein, “Textural features [35] ——, “Artificial life: a new approach to texture classification,” Interna-
for image classification,” IEEE Transactions on Systems, Man and tional Journal of Pattern Recognition and Artificial Intelligence, vol. 19,
Cybernetics, vol. 3, no. 6, pp. 610–621, 1973. no. 3, pp. 355–374, 2005.
[12] R. M. Haralick, “Statistical and structural approaches to texture,” Pro- [36] L. Shenzhou, W. Xiaoqin, L. Qiang, Z. Xiaohui, A. K. Jonathan,
ceedings of the IEEE, vol. 67, no. 5, pp. 786–804, 1979. U. Neha, F. Omenetto, and D. L. Kaplan, “Insoluble and flexible silk
[13] D. Chetverikov, “Texture analysis using feature based pairwise interac- films containing glycerol,” Biomacromolecules, vol. 11, no. 1, pp. 143–
tion maps,” Pattern Recognition, vol. 32, no. 3, pp. 487–502, March 150, November 2010.
1999. [37] B. B. Machado, W. N. Goncalves, and O. M. Bruno, “Artificial crawler
[14] T. Ojala, M. Pietikäinen, and T. Mäenpää, “Multiresolution gray-scale model for texture analysis on silk fibroin scaffolds,” Computational
and rotation invariant texture classification with local binary patterns,” Science and Discovery, vol. 0, no. 7, p. 015004, 2014.
IEEE Trans. Pattern Anal. Mach. Intell., vol. 24, no. 7, pp. 971–987, [38] B. B. Machado, W. N. Goncalves, M. do Santos, and J. F. R. Jr.,
Jul. 2002. [Online]. Available: http://dx.doi.org/10.1109/TPAMI.2002. “Multiscale fractal description using non-liner diffusion of perona-malik
1017623 for texture analysis,” Pattern Recognition Letters, 2016.
[15] X. Tan and B. Triggs, “Enhanced local texture feature sets for face [39] Y. Meyer, Oscillating Patterns in Image Processing and Nonlinear
recognition under difficult lighting conditions,” Image Processing, IEEE Evolution Equations: The Fifteenth Dean Jacqueline B. Lewis Memorial
Transactions on, vol. 19, no. 6, pp. 1635–1650, June 2010. Lectures. Boston, MA, USA: American Mathematical Society, 2001.
[16] A. Hadid, J. Ylioinas, M. Bengherabi, M. Ghahramani, and [40] B. B. Machado, W. N. Goncalves, and O. M. Bruno, “Enhancing the
A. Taleb Ahmed, “Gender and texture classification: a comparative texture attribute with partial differential equations: a case of study
analysis using 13 variants of local binary patterns,” Pattern Recognition with gabor filters,” in Proceedings of the 13th international conference
Letters, vol. 68, no. 2, pp. 231–238, 2015. [Online]. Available: on Advanced concepts for intelligent vision systems, ser. ACIVS’11.
http://www.sciencedirect.com/science/article/pii/S0167865515001348 Berlin, Heidelberg: Springer-Verlag, 2011, pp. 337–348.
[17] G. R. Cross and A. K. Jain, “Markov random field texture models,” [41] B. B. Machado, W. N. Goncalves, M. dos Santos, and J. F. R. Jr., “Iden-
IEEE Trans. Pattern Anal. Mach. Intell., vol. 5, pp. 25–39, 1983. tification of soybean leaf diseases using multiscale fractal descriptiors,”
[18] R. Chellappa and S. Chatterjee, “Classification of textures using gaussian Computer Electronics and Agriculture, 2016.
markov random fields,” IEEE Transactions on Acoustics, Speech, and [42] B. B. Machado, J. Orue, M. dos Santos, D. Sarath, G. Goncalves,
Signal Processing, vol. 33, no. 1, pp. 959–963, August 1985. W. N. Goncalves, H. Pistori, R. R. Mauro, and J. F. R. Jr., “Bioleaf:
[19] B. B. Mandelbrot, Fractals: form, chance, and dimension, ser. Mathe- a professional mobile application to measure foliar damage caused by
matics Series. San Francisco (CA, USA): W. H. Freeman, 1977. insect herbivory,” Computer Electronics and Agriculture, 2016.
[20] C. Tricot, Curves and fractal dimension. Springer-Verlag, 1995. [43] B. B. Machado, L. Scabini, M. do Santos, W. N. Goncalves, R. Moraes,
[21] B. B. Mandelbrot, The Fractal Geometry of Nature. New York: W. H. and J. F. R. Jr., “A complex network approach for nanoparticle agglom-
Freeman and Company, August 1983. eration analysis in nanoscale images,” Information Sciences, 2016.