Local binary patterns

Studies in Computational Intelligence 506 Sheryl Brahnam Lakhmi C Jain Loris Nanni Alessandra Lumini Editors Local Binary Patterns: New Variants and Applications www.it-ebooks.info Studies in Computational Intelligence Volume 506 Series Editor J Kacprzyk, Warsaw, Poland For further volumes: http://www.springer.com/series/7092 www.it-ebooks.info Sheryl Brahnam Lakhmi C Jain Loris Nanni Alessandra Lumini • • Editors Local Binary Patterns: New Variants and Applications 123 www.it-ebooks.info Editors Sheryl Brahnam Computer Information Systems Missouri State University Springfield USA Loris Nanni Departimento di Elettronica e Informatica Università di Padova Padua Italy Lakhmi C Jain Faculty of Education, Science, Technology and Mathematics University of Canberra Canberra Australia Alessandra Lumini Department of Electronic, Informatics and Systems Università di Bologna Cesena Italy ISSN 1860-949X ISBN 978-3-642-39288-7 DOI 10.1007/978-3-642-39289-4 ISSN 1860-9503 (electronic) ISBN 978-3-642-39289-4 (eBook) Springer Heidelberg New York Dordrecht London Library of Congress Control Number: 2013945803 Ó Springer-Verlag Berlin Heidelberg 2014 This work is subject to copyright All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed Exempted from this legal reservation are brief excerpts in connection with reviews or scholarly analysis or material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the Publisher’s location, in its current version, and permission for use must always be obtained from Springer Permissions for use may be obtained through RightsLink at the Copyright Clearance Center Violations are liable to prosecution under the respective Copyright Law The use of general descriptive names, registered names, trademarks, service marks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made The publisher makes no warranty, express or implied, with respect to the material contained herein Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com) www.it-ebooks.info Foreword Texture is an important characteristic of many types of images It can be seen in images ranging from multispectral remotely sensed data to microscopic images Texture can play a key role in a wide variety of applications of computer vision and image analysis Therefore, the analysis of textures has been a topic of intensive research since the 1960s Most of the proposed methods have not been, however, capable to perform well enough for real-world textures In recent years, discriminative and computationally efficient local texture descriptors have been developed, such as local binary patterns (LBP), which has led to significant progress in applying texture methods to different problems and applications The focus of research has broadened from 2D textures to 3D textures and spatiotemporal (dynamic) textures Due to this progress, the division between texture descriptors and more generic image or video descriptors has been disappearing The original LBP operator was invented already two decades ago, but at that time one could not imagine what a great success it would be today In the 1990s it was difficult to get LBP-related papers accepted for leading journals or conferences, because LBP was regarded as an ad hoc method with no theoretical foundation However, the promising power of LBP was already known to some of us, because it performed much better in texture classification and segmentation tasks than the state of the art at that time Due to its computational simplicity and good performance, LBP was also successfully used in some applications such as industrial inspection To the large extent the scientific community found LBP after its generalized version was published in IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) journal in 2002 Many of the leading computer vision scientists became really interested in LBP after it was shown to be highly successful in face recognition, published first at the ECCV 2004 conference and then in the IEEE PAMI journal in 2006 Different types of applications of LBP to motion analysis have been proposed after the spatiotemporal LBP was introduced in 2007, also in IEEE PAMI Due to its discriminative power and computational simplicity, the LBP texture operator has become a highly popular approach in various applications, including, for example, facial image analysis, biometrics, medical image analysis, motion and activity analysis, and content-based retrieval from image or video databases v www.it-ebooks.info vi Foreword Today, the interest in LBP is growing further LBP is not seen just as a simple texture operator, but forms the foundation for a new direction of research dealing with local binary image and video descriptors Many different variants of LBP have been proposed to improve its robustness, and increase its discriminative power and applicability to different types of problems The first book on LBP was published in 2011, the first international workshop was held in 2012, and the first special issue of a journal on LBP will appear in late 2013 Due to this progress, the publication of this edited book is very timely The editors of the book have put together an excellent collection of ten peer-reviewed chapters, covering LBP theory, new LBP variants, and LBP and new variants applied to face recognition I consider this book as a very important resource for researchers, engineers, and graduate students interested in methods and applications of computer vision, image analysis, and pattern recognition Matti Pietikäinen Department of Computer Science and Engineering University of Oulu Oulu, Finland www.it-ebooks.info Preface This book introduces Local Binary Patterns (LBP), arguably one of the most powerful texture descriptors, and LBP variants This volume provides the latest reviews of the literature and a presentation of some of the best LBP variants by researchers at the forefront of textual analysis research and research on LBP descriptors and variants The value of LBP variants is illustrated with reported experiments using many databases representing a diversity of computer vision applications in medicine, biometrics, and other areas There is also a chapter that provides an excellent theoretical foundation for texture analysis and LBP in particular A special section focuses on LBP and LBP variants in the area of face recognition, including thermal face recognition This book will be of value to anyone already in the field as well as to those interested in learning more about this powerful family of texture descriptors Springfield, USA Canberra, Australia Cesena, Italy Padua, Italy Sheryl Brahnam Lakhmi C Jain Alessandra Lumini Loris Nanni vii www.it-ebooks.info Contents Introduction to Local Binary Patterns: New Variants and Applications Sheryl Brahnam, Lakhmi C Jain, Alessandra Lumini and Loris Nanni Part I LBP Theory A Unifying Framework for LBP and Related Methods Francesco Bianconi and Antonio Fernández Part II 17 New LBP Variants Local Phase Quantization for Blur Insensitive Texture Description Janne Heikkilä, Esa Rahtu and Ville Ojansivu 49 The Geometric Local Textural Patterns (GLTP) S A Orjuela Vargas, J P Yañez Puentes and W Philips Local Configuration Features and Discriminative Learnt Features for Texture Description Yimo Guo, Guoying Zhao and Matti Pietikäinen 113 Heterogeneous Ensemble of Classifiers for Sub-Cellular Image Classification Based on Local Ternary Patterns Loris Nanni, Michelangelo Paci and Stefano Severi 131 FLBP: Fuzzy Local Binary Patterns Stamos Katsigiannis, Eystratios Keramidas and Dimitris Maroulis 85 149 ix www.it-ebooks.info x Contents Ensemble of Local Phase Quantization Variants with Ternary Encoding Loris Nanni, Sheryl Brahnam, Alessandra Lumini and Tonya Barrier Part III LBP and New Variants Applied to Face Recognition State-of-the-Art LBP Descriptor for Face Recognition Chi ho Chan, Josef Kittler and Norman Poh 10 Thermal Face Recognition in Unconstrained Environments Using Histograms of LBP Features Javier Ruiz-del-Solar, Rodrigo Verschae, Gabriel Hermosilla and Mauricio Correa 11 177 Histogram-Tensorial Gaussian Representations and its Applications to Facial Analysis John A Ruiz Hernandez, James L Crowley, Augustin Lux and Matti Pietikäinen Editors www.it-ebooks.info 191 219 245 269 Chapter Introduction to Local Binary Patterns: New Variants and Applications Sheryl Brahnam, Lakhmi C Jain, Alessandra Lumini and Loris Nanni Abstract This chapter provides an introduction to Local Binary Patterns (LBP) and important new variants Some issues with LBP variants are discussed A summary of the chapters on LBP is also presented Introduction By the end of this century, computer vision will radically change society Computer vision is already providing machines with an ability to understand their surroundings, to control the quality of products in industrial processes, to diagnose disease, to facilitate factory automation, to self-drive automobiles, to analyze satellite and aerial imagery, to perform document image analysis, to identify people, to recognize gestures, and to synthesize images for computer animation and graphics applications [1] The full extent of the changes about to take place is unknown, but we can expect that future generations will find it as difficult to imagine what life was like before S Brahnam (B) Computer Information Systems, Missouri State University, 901 S National, Springfield, MO 65804, USA e-mail: sbrahnam@missouristate.edu L C Jain University of Canberra, ACT 2601 , Australia e-mail: Lakhmi.jain@unisa.edu.au A Lumini Department of Computer Science and Engineering (DISI), Università di Bologna,Via Venezia 52, 47023 Cesena, Italy e-mail: alessandra.lumini@unibo.it L Nanni Departimento di Elettronica e Informatica (DEI), Università di Padova, Via Gradenigo 6, 35131 Padua, Italy e-mail: nanni@dei.unipd.it S Brahnam et al (eds.), Local Binary Patterns: New Variants and Applications, Studies in Computational Intelligence 506, DOI: 10.1007/978-3-642-39289-4_1, © Springer-Verlag Berlin Heidelberg 2014 www.it-ebooks.info 257 (a) (b) 0.9 0.95 0.88 Recognition Rate Recognition Rate 11 Histogram-Tensorial Gaussian Representations 0.86 0.84 T F 0.82 0.8 0.78 0.9 0.85 0.8 0.75 10 T F (c) (d) 0.8 0.75 0.75 0.7 T F 0.65 0.6 6 10 10 Rank Recognition Rate Recognition Rate Rank 10 0.7 0.65 T F 0.6 0.55 Rank Rank Fig Rank-1 recognition accuracy versus dimensionality reduction with MPCA for the vectors yT and y F using Mag, LoG and γ on the four FERET probe sets (a) f b (b) f c , (c) Dup I and (d) Dup I I size are highlighted in bold and compared with the learning-based face recognition methods Age Estimation Faces are an important source of social and behavioural information, within this information, it is the special interest the ageing information, for its potential applications in (HCI) [23, 34], law-enforcement applications [33, 34], video surveillance [24, 34], appearance prediction across aging [37], and targeting of publicity [17] Despite the number of potential applications, automatic image-based age estimation remains a challenging problem Compared with other facial variations, ageing effects are very dependent on genetics [18], life style, location of residence [19] and weather conditions [24] Furthermore, males and females age differently, and the apparent effects of aging are often masked by make-up and facial accessories [37] Accommodating the influence of individual differences to provide a general method for estimating age based on facial images remains an open problem www.it-ebooks.info 258 J A Ruiz Hernandez et al Table The Rank-1 average recognition error rates on the Yale B + Extended Yale B dataset with different training set sizes Method Train Set Size 10 20 30 ORO [20] SR [4] RDA [4] KLPPSI [2] CTA [11] Eigenfaces [11] Fisherfaces [11] Laplacianfaces [11] Volterrafaces (Linear) [22] Volterrafaces (Quad) [22] y F (Mag, Log, γ) + KDCV (q = 28322) yT (Mag, Log, γ) + KDCV (q = 18818) 24.74 16.99 54.73 37.56 34.08 6.35 13.0 6.94 6.61 12.0 11.6 9.93 7.60 36.06 18.91 18.03 2.67 3.98 1.12 0.90 4.7 4.2 3.15 4.96 31.22 16.87 30.26 0.90 1.27 0.36 0.32 9.0 2.0 1.8 1.39 2.94 27.71 14.94 20.20 0.42 0.58 0.33 0.32 Method Train Set Size 58.0 54.0 26.23 18.23 40.81 20.47 32.18 16.46 32.42 16.26 50 9.33 14.42 10.08 9.73 40 1.0 0.9 0.34 0.43 0.17 0.19 MLASSO [31] SR [4] RDA [4] Volterrafaces (Linear) [22] Volterrafaces (Quad) [22] y F (Mag, Log, γ) + KDCV (q = 28322) yT (Mag, Log, γ) + KDCV (q = 18818) Most automatic image-based age estimation systems are composed by combining two components [24]: an image representation and an age estimation process The age estimation process can be formulated as a multi-classification problem [23] where each age is considered a separate class, a regression problem where an approximative age-function is computed from a set of training images [17, 19] or an hybrid version that combines classification and regression methods [18] In this part of the chapter age estimation problem has been addressed using the tensorial representations explained in Sects and 4, finally Relevance Vector Machines are used as regressor to estimate the age of the subject on the image In age estimation from faces, HBGM (see Sect 2) provides a robust facial representation capable of encoding ageing information in two ways: in appearance using Binary Gaussian Maps and in shape using tensorial representations, both of them combined using Multilinear principal Component Analysis, provide a robust facial feature www.it-ebooks.info 11 Histogram-Tensorial Gaussian Representations 259 6.1 Age Estimation Using RVM To estimate age from faces, we have used the vectors yT and y F In both cases the resulting vector, after applying MPCA, was used as input to train a RVM regressor When a candidate facial image is present, first we apply our tensorial transformation mentioned in Sect 3, then once the final vectors are computed, we try to determine its correct age using the learned regression function In the following sections, we are going to show that our tensorial representations can encode the necessary facial information for determining the age of a subject using the trained regression function In all of our experiments, the images were cropped using manually located eye positions and normalized in size to 64 × 64 pixels The Binary Gaussian Receptive √ Maps √ are calculated in a half-octave Gaussian pyramid with four levels (σ = 2, 2, 2 and 4) A border of pixels in each pyramid level is left untested for faces to avoid problems related with image borders To calculate histograms, we used a sub-region size of × pixels, removing two bins corresponding to values of and 255 per each histogram The remaining bins are grouped to form a 127 bin histogram We used the publicly available Matlab® implementation of the RVM algorithm provided by [39].2 After applying MPCA, each final vector is normalized to unit standard deviation For the RVM algorithm we used a Gaussian kernel k(x, y) = e− x−y /q with a scale parameter q determined using a tuning dataset chosen randomly from the training dataset (the value of q is reported in each experiment and is the only one fixed by tuning) 6.2 Experimental Datasets We have performed several experiments to compare different approaches for estimating age from facial images Two publicly available databases have been used in our experiments: The FG-NET database3 and the MORPH [35] database The performance of age estimation is measured by the mean absolute error (MAE) and the cumulative score (CS) • The MAE is defined as the average of the absolute errors between the estimated N ˆ ages and the ground truth ages M AE = k=1 | Agek − Agek |/N , where Agek ˆ is the ground truth age for the test image k, Agek is the estimated age, and N is the total number of test images MAE is only an indicator of average performance for age estimators, it does not provide enough information of how accurate the estimators might be • The accuracy can be estimated by the cumulative score (CS) that is defined as C S ( j) = Ne≤ j /N × 100, where Ne≤ j is the number of test images on which the estimator makes an absolute error non-higher than j years http://www.vectoranomaly.com/downloads/downloads.htm The FG-NET ageing database, http://www.fgnet.rsunit.com/ www.it-ebooks.info 260 J A Ruiz Hernandez et al 6.2.1 FG-NET Ageing Dataset The FG-NET (Face and Gesture Recognition Research) Ageing Database contains 1,002 face images of 82 subjects from multiple races with age ranges from to 69 years Each image in the database has 68 labelled facial landmarks characterizing shape features not used in our approach Since the images were retrieved from reallife albums of different subjects, aspects as illumination, head pose, facial expressions etc are uncontrolled in this dataset Leave-One-Person-Out (LOPO) mode is used for testing our approaches in the FG-NET database In this mode, the images of one person are used as the test set and those of the others are used as the training set After 82 folds each subject has been used as a the test set once and the final results are calculated based in the result of each fold 6.2.2 MORPH Aging Dataset The MORPH Ageing Database contains 1,724 face images of 515 subjects In our experiment with this database, we use the same testing protocol used by [13] The images on these dataset are only used to test the algorithms trained on the FG-NET database In addition, because all subjects in the FG-NET database are Caucasian descent, only the 433 images of Caucasian descent in the MORPH database are used as the test set 6.3 Comparing yT and yF Performances Our first experiment investigates the performance of each type of configuration for the age estimation problem We compared them on the FG-NET database and we report the results in Table Different orientations and Gaussian derivative orders were tested to obtain the best configuration We used orientations between and π and derivative orders up to fourth, giving the next bank of Gaussian derivative filters used in our age estimation experiments for computing the Histograms of Gaussian Table MAEs on the FG-NET database for different tensorial configurations Orientations θ = {0 : π/5 : π} θ = {0 : π/7 : π} θ = {0 : π/5 : π} θ = {0 : π/7 : π} yT + RVM n = {1, 2, 3} n = {1, 2, 3, 4} n = {3, 4} 5.25 (q = 28.72) 5.16(q = 30.82) 5.23 (q = 50.00) 5.18 (q = 50.00) 5.25 (q = 29.58) 5.23 (q = 30.82) 5.17 (q = 66.33) 4.96(q = 66.33) 5.53 (q 5.48 (q 5.49 (q 5.46 (q www.it-ebooks.info = 29.58) = 30.82) = 38.73) = 38.73) 11 Histogram-Tensorial Gaussian Representations 261 Binary Maps (HGBM) √ √ F(n,θ∈[0,π]) = G n,θ x, y, σ = { 2, 2, 2, 4} with n = {1, 2, 3, 4} (7) Using this filter bank, we are able to compute four different tensors: Tn ∈ R32×orientations×127 n = {1, 2, 3, 4} (8) Experimental results show that the best performance can be achieved with orientations (0, π/7, 2π/7, 3π/7, 4π/7, 5π/7, 6π/7andπ) and four derivative orders, organized in a y F configuration, the second best result was achieved with orientations (0, π/5, 2π/5, 3π/5, 4π/5andπ) and three derivative orders in a yT configuration The best results were highlighted in the Table and used in the following experiments 6.4 Results in the FG-NET Database We compared the two best tensor configurations of Table with the most relevant results of the state-of-the-art in age estimation In Table 5, we report the results for seven different age groups between and 69 years In this table we observed that our method outperforms other methods for age groups 40–59 and 20–29 years For other age groups, our method is competitive and sometimes superior with competing approaches in age estimation The experiments the robustness of our method to textural changes that occur in the periods from adulthood to old age In the FG-NET database, the MAEs of our method are 5.16 for the yT configuration and 4.96 for the y F configuration Comparisons with alternative approaches are reported in Table Our results are comparable with the latest state-of-the-art methods in automatic age estimation in the FG-NET database Comparisons of cumulative scores (CS) on the FG-NET database are shown in Fig 9a We can observe that despite the MAE results for our methods, our approaches outperforms state-of-the-art methods in low age error levels (Err or level ≤ 4), with almost % of improvement in accuracy C S≤4 = 73 %, in addition for high error levels our method has an C S≤10 = 88 % similar to BIF [19] (C S≤10 = 89 %) that as far as we know the best result in the FG-NET dataset 6.5 Results in the MORPH(test) Database More experiments were conducted on the MORPH(test) aging database From the results in the Table 7, the MAEs results of our method are 6.19 and 6.77 for y F and yT respectively, those results outperforms the AGES method [13] in almost two years of www.it-ebooks.info 262 J A Ruiz Hernandez et al Table MAE (years) at different age groups on FG-NET Range # img 0–9 10–19 20–29 30–39 40–49 50–59 60–69 Total 371 339 144 70 46 15 1002 Range 0–9 10–19 20–29 30–39 40–49 50–59 60–69 Total # img 371 339 144 70 46 15 1002 Method yT + RVM 3.14 4.05 4.72 10.08 14.15 22.06 33.12 5.16 Method RUN [43] 2.51 3.76 6.38 12.51 20.09 28.07 42.50 5.78 y F + RVM BIF [19] 3.19 3.90 4.29 9.17 13.76 20.06 32.25 4.96 2.99 3.39 4.30 8.24 14.98 20.49 31.62 4.77 QM [23] 6.26 5.85 7.10 11.56 14.80 24.27 37.38 7.57 MLP [23] 11.63 3.33 8.81 18.46 27.98 49.13 49.13 10.39 Table MAE (years) comparisons on FG-NET Method MAE (Years) QM [23] MLPs [23] RUN [43] BM [45] LARR [17] PFA [18] BIF [19] yT + RVM y F + RVM 6.55 6.98 5.78 5.33 5.07 4.97 4.77 5.16 4.96 difference and other methods like SVM and WAS with a difference of almost three years The CS curves on the MORPH database are shown in Fig 9b Our method outperforms other methods in error levels for all of the age groups with an C S≤10 = 84 % against the AGESlda method [13] with a C S≤10 = 78 % www.it-ebooks.info 11 Histogram-Tensorial Gaussian Representations 263 Table MAE (years) comparisons on MORPH (Test Set) Method MAE (Years) WAS [13] SVM [13] AGES [13] AGESlda [13] yT + RVM y F + RVM 9.32 9.23 8.83 9.32 6.77 6.19 (b) Cumulative Score(%) 90 80 70 60 50 T F 40 BIF BM MLP QM 30 20 10 0 10 11 12 13 14 15 16 17 18 19 20 Cumulative Score(%) (a) 100 90 80 70 60 50 40 F T 30 WAS SVM 20 10 0 Error Level(Years) 10 Error Level(Years) Fig Cumulative scores on a FG-NET database and b MORPH (test) database Conclusions We have introduced a new image representation model that uses a simple set of Gaussian filters calculated effectively with a Half-Octave Gaussian pyramid and a tensorial representation that conserves the natural structure of the feature space described for such filters Two algorithmic structures for fusing tensors using MPCA have been proposed Each one of these structures have been applied to solve the problems of face recognition and age estimation, showing with the results, the performance of these representations to describe human facial appearance Appendix Gaussian Scale Space The term Scale Space was introduced by Witkin [41] to describe the blurring properties of one dimensional signals Koenderink and van Doom [21] applied this concept to images using the assumptions of causality, isotropy and homogeneity for revealing that the scale space must be essentially governed by the isotropic diffusion equation d I /dσ = c∇ I which shows that many physical phenomena can be described using www.it-ebooks.info 264 J A Ruiz Hernandez et al the Gaussian kernel: G(x, y, σ) = − x +y2 e 2σ 2σπ (9) where σ is the size of the support in terms of the second moment (or variance) Crowley and Stern [9]; Crowley and Sanderson [8]; Burt and Adelson [3] present the first notions for computing the scale space using a pyramidal representation and finally, Lindeberg [26] formalized the concept of the discrete Gaussian scale space Following the above mentioned, the Gaussian scale space can be computed as follows: I (x, y, σ) = G(x, y, σ) ∗ I (x, y) = − x +2y e 2σ ∗ I (x, y) 2σπ (10) where I (x, y, σ) is the Gaussian scale-space representation of the image I (x, y) and "∗" is the convolution operator The Gaussian Derivatives In Neuroscience, the classical receptive field of a sensory neuron is a region of space in which the presence of a stimulus will alter the firing of that neuron For mammals, receptive fields have been identified for neurons of the auditory system, the somatosensory system, and the visual system Young et al [47] have reported that receptive fields in the visual cortex can be well modelled using Gaussian derivative operators up to fourth order To describe Gaussian derivatives, we introduce a particular notation which will be used to describe the derivatives in this chapter Let v(θ) = (cos(θ) sin(θ)), be the directional vector that describes the desired orientation θ for a Gaussian derivative of nth order In addition, we define the x-axis parallel to v(0◦ ), which corresponds to θ = The y-axis is defined by θ = 90 ◦ and is therefore parallel to v(90 ◦ ) Following this notation, Gaussian derivatives of nth order at any orientation θ are described by: ∂n (11) G n,θ (x, y, σ) = n G(x, y, σ) ∂ v 9.1 Filters Based in Gaussian Derivatives Based in Gaussian derivatives up to the second order, it is possible to compute a set of invariant filters which are invariant to illumination variations and can be easily computed using the Half-Octave Gaussian Pyramid algorithm This filter set has been used with success in texture, object and pixel classification [7, 16, 25] www.it-ebooks.info 11 Histogram-Tensorial Gaussian Representations Mag(x, y, σ) = 265 (G x (x, y, σ))2 + G y (x, y, σ) LoG(x, y, σ) = G x (x, y, σ) + G y (x, y, σ) γ(x, y, σ) = σ G 2x (x, y, σ) − G 2y (x, y, σ) + G x y (x, y, σ)2 (12) (13) (14) where Mag corresponds to the gradient magnitude which is invariant to rotation changes and is computed with Gaussian derivatives of first order, LoG is the well known Laplacian of Gaussians operator, proposed by [26] and γ is the third component of the second local order Gaussian jet norm, proposed by [15] The main advantage of these filters is their invariance to strong changes in illumination present in facial analysis, specially in the face recognition application 9.2 The Gaussian Jet Koenderink and van Doom [21] argue that the local visual appearance in an image neighbourhood can be represented by a local Taylor series expansion of the neighbourhood, computed using local Gaussian derivatives The coefficients of this series constitute a feature vector, referred to as the “Local Jet” that compactly represents image appearance and can be used for indexing, matching and recognition Romeny et al [36] have shown that invariance to scale and orientation can be obtained when the local jet is computed using Gaussian derivatives It be I x m y n (x, y, σ) an image filtered with a Gaussian derivative: ∂m + n G(x, y, σ) ∗ I (x, y) ∂ m x∂ n y ∂m + n = m n (G(x, y, σ) ∗ I (x, y)) ∂ x∂ y ∂m + n = m n I (x, y, σ) ∂ x∂ y I x m y n (x, y, σ) = (15) From the equation above, the Gaussian jet of kth order is defined as follows: jk (x, y, σ) = I (x, y, σ), I x (x, y, σ), I y (x, y, σ), , I x m y n (x, y, σ) jk (x, y, σ) ∈ (k + 1)(k + 2) www.it-ebooks.info with m + n = k (16) 266 J A Ruiz Hernandez et al The Second Local Order Gaussian Jet Norm A mathematical definition of norm for a second order Gaussian jet (k = 2) has been proposed by Griffin [15] This norm is defined as the minimum of scale space norms from a set of profiles measured by the jet at a given point in the image and is defined as: j2 (x, y, σ) = σ I x2 + I y2 + 41 σ I 22 + I 22 + 41 σ x y Ix − I y2 + 4I x2y (17) The Second Order Gaussian Jet norm satisfies all the mathematical characteristics of a jet norm In particular, it is unaffected by adding a constant to image intensities [15] and thus can avoid problems due to changes in illumination References Ahonen, T., Hadid, A., Pietikainen, M.: Face description with local binary patterns: Application to face recognition IEEE Transact Pattern Anal.Mach.Intell 28(12), 2037–2041 (2006) An, Senjian, Liu, Wanquan: and Svetha Venkatesh Exploiting side information in locality preserving projection, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2008) Burt, P., Adelson, E.: The laplacian pyramid as a compact image code IEEE Transactions on Communications 31(4), 532–540 (1983) Cai, Deng, He, Xiaofei: and Jiawei Han, pp 1–8 Spectral regression for efficient regularized subspace learning, Proceedings of the IEEE International Conference on Computer Vision (2007) Cevikalp, H., Neamtu, M., Wilkes, M.: Discriminative common vector method with kernels IEEE Transactions on Neural Networks 17(6), 1550–1565 (2006) Cevikalp, H., Neamtu, M., Wilkes, M., Barkana, A.: Discriminative common vectors for face recognition IEEE Transactions on Pattern Analysis and Machine Intelligence 27(1), 4–13 (2005) M Crosier and L.D Griffin, Texture classification with a dictionary of basic image features, Proceedings of the IEEE Conference on Computer Vision and, Pattern Recognition, 2008, pp 1–7 Crowley, James L., Sanderson, Arthur C.: Multiple resolution representation and probabilistic matching of 2-d gray-scale shape IEEE Transactions on Pattern Analysis and Machine Intelligence 9(1), 113–121 (1987) James, L.: Crowley and R.M Stern, Fast computation of the difference of low-pass transform IEEE Transactions on Pattern Analysis and Machine Intelligence 6(2), 212–222 (1984) 10 Tao Dacheng, Li Xuelong, Wu Xindong, and S.J Maybank, General tensor discriminant analysis and gabor features for gait recognition, IEEE Transactions on Pattern Analysis and Machine Intelligence 29 (2007), no 10, 1700–1715 11 Yun Fu and T S Huang, Image classification using correlation tensor analysis, IEEE Transactions on Image Processing 17 (2008), no 2, 226–234 12 Geng, X., Smith-Miles, K., Zhou, Z.-H., Wang, L.: Face image modeling by multilinear subspace analysis with missing values IEEE Transactions on Systems, Man, and Cybernetics 41(3), 881–892 (2011) www.it-ebooks.info 11 Histogram-Tensorial Gaussian Representations 267 13 Geng, X.: Zhi-Hua Zhou, and K Smith-Miles, Automatic age estimation based on facial aging patterns, IEEE Transactions on Pattern Analysis and Machine Intelligence 29(12), 2234–2240 (2007) 14 Georghiades, Athinodoros S., Belhumeur, Peter N., Kriegman, David J.: From few to many: Illumination cone models for face recognition under variable lighting and pose IEEE Transactions on Pattern Analysis and Machine Intelligence 23(6), 643–660 (2001) 15 Griffin, L.D.: The second order local-image-structure solid IEEE Transactions on Pattern Analysis and Machine Intelligence 29(8), 1355–1366 (2007) 16 Lewis D Griffin, Martin Lillholm, Michael Crosier, and Justus van Sande, Basic image features (bifs) arising from approximate symmetry type, Proceedings of the Second International Conference on Scale Space and Variational Methods in Computer Vision, 2009, pp 343–355 17 Guo, G., Fu, Y., Dyer, C.R., Huang, T.S.: Image-based human age estimation by manifold learning and locally adjusted robust regression IEEE Transactions on Image Processing 17(7), 1178–1188 (2008) 18 G Guo, Y Fu, Charles R Dyer, and Thomas S Huang, A probabilistic fusion approach to human age prediction, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2008 19 G Guo, G Mu, Y Fu, and Thomas S Huang, Human age estimation using bio-inspired features., Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp 112–119 20 Hua, Gang, Viola, Paul A.: and Steven M Drucker, Face recognition using discriminatively trained orthogonal rank one tensor projections, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2007) 21 J.J Koenderink and A.J van Doom, Representation of local geometry in the visual system, Biol Cybern 55 (1987), no 6, 367–375 22 Kumanr, R., Banerjee, A., Vemuri, B.C.: Volterrafaces: Discriminant analysis using volterra kernels Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, June (2009) 23 A Lanitis., C Draganova., and C Christodoulou, Comparing different classifiers for automatic age estimation, IEEE Transactions on Systems, Man, and Cybernetics 34 (2004), no 1, 621– 628 24 Lanitis, A., Taylor, C.J., Cootes, T.F.: Toward automatic simulation of aging effects on face images IEEE Transactions on Pattern Analysis and Machine Intelligence 24(4), 442–455 (2002) 25 Lillholm, M., Griffin, L.: Novel image feature alphabets for object recognition Proceedings of the International Conference on Pattern Recognition, Dec (2008) 26 Lindeberg, T.: Scale-space theory in computer vision Kluwer Academic Publishers, Norwell, MA, USA (1994) 27 Lu, H., Plataniotis, K.N., Venetsanopoulos, A.N.: Mpca: Multilinear principal component analysis of tensor objects IEEE Transactions on Neural Networks 19(1), 18–39 (2008) 28 Yui Man Lui and J Ross Beveridge, Grassmann registration manifolds for face recognition, Proceedings of the European Conference on Computer Vision, 2008, pp 44–57 29 Ojala, T.: M Pietikinen, and Menp T., Multiresolution gray-scale and rotation invariant texture classification with local binary patterns IEEE Transactions on Pattern Analysis and Machine Intelligence 24(7), 971–987 (2002) 30 Ojala, Timo: M Pietikinen, and D Harwood, A comparative study of texture measures with classification based on featured distributions, Pattern Recognition 29, 51–59 (1996) 31 Duc-Son Pham and Svetha Venkatesh, Robust learning of discriminative projection for multicategory classification on the stiefel manifold, Proceedings of the IEEE Conference on Computer Vision and, Pattern Recognition, 2008 32 Phillips, P.J., Moon, H.: Syed A Rizvi, and Patrick J Rauss, The feret evaluation methodology for face-recognition algorithms IEEE Transactions on Pattern Analysis and Machine Intelligence 22(10), 1090–1104 (2000) www.it-ebooks.info 268 J A Ruiz Hernandez et al 33 Ramanathan, N., Chellappa, R.: Face verification across age progression IEEE Transactions on Image Processing 15(11), 3349–3361 (2006) 34 Ramanathan, N., Chellappa, R., Biswas, S.: Computational methods for modeling facial aging: A survey J Vis Lang Comput 20(3), 131–144 (2009) 35 Jr, Karl: Ricanek and Tamirat Tesafaye A longitudinal image database of normal adult ageprogression, Proceedings of the IEEE face and gesture recognition, Morph (2006) 36 Bart M ter Haar Romeny, Luc Florack, Alfons H Salden, and Max A Viergever, Higher order differential structure of images, Proceedings of the 13th International Conference on Information Processing in Medical, Imaging, 1993, pp 77–93 37 Suo, Jinli, Zhu, Song-Chun, Shan, Shiguang, Chen, Xilin: A compositional and dynamic model for face aging IEEE Transactions on Pattern Analysis and Machine Intelligence 32(3), 385–401 (2010) 38 X Tan and B Triggs, Fusing gabor and lbp feature sets for kernel-based face recognition, Proceedings of the 3rd international conference on Analysis and modeling of faces and gestures, AMFG’07, 2007, pp 235–249 39 Michael, E.: Tipping Sparse bayesian learning and the relevance vector machine, Journal of Machine Learning Research 1, 211–244 (2001) 40 M Turk and A Pentland, Face recognition using eigenfaces, Proceedings of the IEEE Conference on Computer Vision and, Pattern Recognition, 1991, pp 586–591 41 A Witkin, Scale-space filtering: A new approach to multi-scale description, IEEE International Conference on Acoustics, Speech, and, Signal Processing., vol 9, mar 1984, pp 150–153 42 Wu, J., Rehg, J.M.: Where am i: Place instance and category recognition using spatial pact, pp 1–8 Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, June (2008) 43 Yan, S., Wang, H., Tang, X.: and T S Huang, Learning auto-structured regressor from uncertain nonnegative labels, Proceedings of the IEEE International Conference on Computer Vision (2007) 44 Yan, S., Xu, D., Yang, Q., Zhang, L., Tang, X., Zhang, H.J.: Multilinear discriminant analysis for face recognition IEEE Transactions on Image Processing 16(1), 212–220 (2007) 45 Yan, Shuicheng, Wang, Huan, Tang, Xiaoou, Liu, Jianzhuang, Huang, Thomas S.: Regression from uncertain labels and its applications to soft biometrics IEEE Transactions on Information Forensics and Security 3(4), 698–708 (2008) 46 Yang, Jian, Zhang, David, Frangi, Alejandro F.: and Jing yu Yang Two-dimensional pca: A new approach to appearance-based face representation and recognition, IEEE Transactions on Pattern Analysis and Machine Intelligence 26(1), 131–137 (2004) 47 Richard A Young, Ronald M Lesperance, and W Weston Meyer, The gaussian derivative model for spatial-temporal vision: I cortical model., Spatial Vision 2001 (2001), 3–4 48 Zhang, B., Shan, S., Chen, X., Gao, W.: Histogram of gabor phase patterns (hgpp): A novel object representation approach for face recognition IEEE Transactions on Image Processing 16(1), 57–68 (2007) 49 W Zhang, S Shan, W Gao, X Chen, and H Zhang, Local gabor binary pattern histogram sequence (lgbphs): A novel non-statistical model for face representation and recognition, Proceedings of the IEEE International Conference on Computer Vision, 2005 www.it-ebooks.info Editors Sheryl Brahnam is the Director/Founder of Missouri State University’s infant COPE (Classification Of Pain Expressions) project Her interests focus on face recognition, face synthesis, medical decision support systems, embodied conversational agents, computer abuse, and artificial intelligence She has worked extensively with Professors Lakhmi C Jain, Loris Nanni, and Alessandra Lumini Dr Brahnam has published articles related to medicine in such journals as Artificial Intelligence in Medicine, Expert Systems with Applications, Journal of Theoretical Biology, Amino Acids, Neural Computing and Applications, and Decision Support Systems Lakhmi C Jain serves as Adjunct Professor at the University of South Australia, Australia and University of Canberra, Australia He is a Fellow of Engineers Australia Dr Jain founded the KES International for providing a professional community the opportunities for publications, knowledge exchange, cooperation and teaming Involving around 5000 researchers drawn from universities and companies world-wide, KES facilitates international cooperation and generate synergy in teaching and research KES S Brahnam et al (eds.), Local Binary Patterns: New Variants and Applications, Studies in Computational Intelligence 506, DOI: 10.1007/978-3-642-39289-4, © Springer-Verlag Berlin Heidelberg 2014 www.it-ebooks.info 269 270 Editors regularly provides networking opportunities for professional community through one of the largest conferences of its kind in the area of KES (http://www.kesinternational org) His interests focus on the artificial intelligence paradigms and their applications in complex systems, security, e-education, e-healthcare, unmanned air vehicles and intelligent agents Alessandra Lumini received her Laurea Degree cum laude in Computer Science from the University of Bologna, Italy, on March 26th 1996 Since 1996 she carried out her research activity at the Dipartimento di Elettronica, Informatica e Sistemistica (DEIS), University of Bologna, Italy, where in 2001 she received her Ph.D for her work on “Image Databases” Since October 2002 she has been an Associate Researcher at II Faculty of Engineering of University of Bologna She currently teaches Information Systems LA and Information Systems LS in the II Faculty of Engineering of University of Bologna, in Cesena She carries out research at DEIS in the fields of biometric systems, pattern recognition, image databases and bioinformatics She has extensively served as a referee for such international journals as IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE Signal Processing Information Forensics and Security, IEEE Transactions on Circuits and Systems for Video Technology, IEEE Transactions on Systems, Man, and Cybernetics, and Pattern Recognition Letters She has coauthored more than 100 research papers She has been involved in several projects supported by MURST and European Community She is member of the Biometric Systems Lab at Computer Science—Cesena Her recent research interests are mainly directed towards the areas of biometric systems and bioinformatics Loris Nanni received his Master Degree cum laude in Computer Science from the University of Bologna, Italy, on June 2002 From 2002 to 2010 he carried out his research activity at the Dipartimento di Elettronica, Informatica e Sistemistica (DEIS), University of Bologna, Italy, where he received his Ph.D degree in 2006 On September, 29th 2006 he won a post PhD fellowship from the university of Bologna (from October 2006 to October 2008), where he ranked first in the competitive examination in the industrial engineering area Loris Nanni was a Post PhD fellow from 2006 to 2008 and from 2010 to 2011 at DEIS In 2009 he worked under the supervision of prof Sheryl Brahnam, Missouri State University, on the creation www.it-ebooks.info Editors 271 of software to recognize and track visual input and modules that utilize artificially intelligent algorithms to process visual output, funded by the Provost’s Futures Grant for Artificial Intelligence in Artistic Process Expression In December, 2010, he won a position as Associate Researcher at the Department of Information Engineering at the University of Padua (Padova) He carries out research in the fields of biometric systems, pattern recognition, image databases and bioinformatics He has extensively served as a referee for several international journals He is author/coauthor of more than 150 research papers He has been involved in some projects supported by MURST and the European Community www.it-ebooks.info ... Center-symmetric local binary patterns Median binary patterns Local ternary patterns Centralized binary patterns Improved local ternary patterns Completed local binary patterns Local quinary patterns Binary. .. hILTPH ||hILTPL 5.14 Completed Local Binary Patterns Completed Local Binary Patterns (CLBP) have been recently introduced by Guo et al [24] as an extension of local binary patterns The approach is... from zero to eight, there are nine possible patterns The kernel function is: f RT (x) = b Ic − I j − (17) j =0 5.5 Local Binary Patterns Local binary patterns (LBP) [59] have received a great deal