FACIAL SOFT BIOMETRICS - Library of Ph.D. Theses | EURASIP
FACIAL SOFT BIOMETRICS - Library of Ph.D. Theses | EURASIP FACIAL SOFT BIOMETRICS - Library of Ph.D. Theses | EURASIP
16 1. INTRODUCTIONsystems, and describe how the employed categorization algorithms (eye color detector, glasses andmoustache detector) are applied on a characteristic database of 646 people. In the same Section wefurthermore provide simulations that reveal the variability and range of the pruning benefits offeredby different SBSs. In Section 4.7 we derive concise closed form expressions on the measures ofpruning gain and goodput, provide simulations, as well as derive and simulate aspects relating tothe complexity costs of different soft biometric systems of interest.Chapter 5 - Frontal-to-side person re-identificationTypically biometric face-recognition algorithms are developed, trained, tested and improvedunder the simplifying assumption of frontal-to-frontal person recognition. Such algorithms thoughare challenged when facing scenarios that deviate from the training setting, such as for examplein the presence of non-constant viewpoints, including the frontal-to-side scenario. Most personrecognition algorithms, whether holistic or based on facial features, only manage to optimally handlepose differences that are less than about15 degrees. As a result, a variation in the pose is oftena more dominant factor than a variation of subjects. This aspect of pose variation comes to the forein video surveillance, where a suspect may be pictured firstly frontal, whereas the correspondingtest images could be captured from the side, thus introducing a frontal-to-side recognition problem.Towards handling this problem, we employ multiple soft biometrics related traits. One ofour tasks here is to get some insight into the significance of these traits, specifically the significanceof using hair, skin and clothes patches for frontal-to-side re-identification. We are workingon the color FERET dataset [Fer11] with frontal gallery images for training, and side (profile)probe images for testing. Towards achieving re-identification, the proposed algorithm first analyzesthe color in Section 5.2.4.1 and furthermore texture in Section 5.2.4.2 of the three patches.Then we study the intensity correlations between patches in Section 5.2.4.3. This analysis is thenfollowed by the construction of a single, stronger classifier that combines the above measures inSection 5.2.5, to re-identify the person from his or her profile.Deviating from the above security related applications, we consider then an application closerto entertainment, and specifically consider the application of soft biometrics in analyzing andquantifying facial aesthetics.Chapter 6 - Soft biometrics for quantifying and predicting facial aestheticsWith millions of images appearing daily on Facebook, Picasa, Flickr, or on different socialand dating sites, photographs are often seen as the carrier of the first and deciding impression of aperson. At the same time though, human perception of facial aesthetics in images is a priori highlysubjective.We related among others soft biometric traits with this subjective human perception. In theprovided study we quantify insight on how basic measures can be used to improve photographsfor CVs or for different social and dating websites. This helps create an objective view on subjectiveefforts by experts / journalists when retouching images. We use the gained objective viewto examine facial aesthetics in terms of aging, facial surgery and a comparison of average femalesrelatively to selected females known for their beauty. Specifically in Section 6.3 we introducethe employed database, as well as describe the basic features and methods used in this study. InSection 6.4 we proceed with numerical results, and provide intuition on the role of features, imagequality and facial features, in human perception. In Section 6.5, we use these accumulated conclusionsto construct a basic linear model that predicts attractiveness in facial photographs usingdifferent facial traits as well as image properties. We then examine and validate the designed met-
17ric. In Section 6.6 we employ the developed metric to conduct experiments and answer questionsregarding the beauty index in three cases: for famous attractive females, for aging females and incase of facial surgery. Finally we proceed to simulate in Section 6.7 based on both, the presentedmetric, as well as state of the art algorithmic accuracies an automatic tool for beauty prediction.Chapter 7 - Practical implementation of soft biometrics classification algorithmsTowards practical implementation of the related concepts and ideas, in Chapter 7 we developa tool (concatenation of classification algorithms) for classification of facial soft biometric traits,where we specifically emphasize on the most obvious facial identifiers, primarily mentioned byhumans, when portraying an unknown individual. The constructed tool is streamlined to achievereliability of identification at reduced complexity, and hence focuses on simple yet robust softbiometrictraits, including hair color, eye color and skin color, as well as the existence of beard,moustache and glasses. We then specifically focus on extraction and categorization of eye color,and present an additional study where we illustrate the influence of surrounding factors like illumination,eye glasses and sensors on the appearance of eye color.In Section 7.1 a bag of six facial soft biometrics is elaborated, for which estimation algorithmsare featured, along with the related experimental results, see Section 7.1.2. We then proceedto focus on eye color as a soft biometric trait in Section 7.2 and examine an automatic eye colorclassifier in challenging conditions, such as changing illumination, presence of glasses and camerasensors, see Section 7.4.Chapter 8 - User acceptance study relating to soft biometricsFinally we conclude with a usability study that verifies the user acceptance of SBSs, specificallywhen compared to existing PIN or fingerprint access control systems.The pervasiveness of biometric systems, and the corresponding growth of the biometric marketsee [usa11a], has successfully capitalized on the strength of biometric-based methods in accuratelyand effectively identifying individuals. As a result, modern state-of-the-art intrusion detection andsecurity systems include by default at least one biometric trait. It is the case though that little emphasishas been given to better understanding user-acceptance and user-preference regarding suchsystems. Existing usability related works, such as in [CAJ03] and [LBCK03], focus on establishingfunctional issues in existing ATM machines, or on studying the influence of user interactionon the performance of fingerprint based systems (see [KED11]) and interfaces (see [RJMAS09]).Other interesting works (see [usa11b], [CG05], [CJMR09]), analyze possible methods that improveinterface design. Our emphasis here is on providing insight on the attitudes and experiencesof users towards novel and emerging biometric verification methods, and to explore whether suchnovel biometric technologies can be, in terms of user acceptance, valid alternatives to existingprevalent PIN based systems. Our focus, in addition to considering the traditional PIN-basedmethod, is to explore the usability aspects of systems based on classical biometrics such as fingerprintand face recognition, and to then proceed to study the usability of systems based on theemerging class of soft-biometric methods. Our evaluation is based on having the users rate andrank their experiences with different access methods.In Section 8.2 we briefly describe the user test setting, as well as the conditions and the performedtest procedures. We then proceed to elaborate on the chosen verification methods and onthe designed interfaces. In Section 8.2 we present the results obtained from the user study, in termsof evaluation and quantification of the different usability measurement characteristics. In the samesection we provide the user test outcomes of direct comparisons between the four presented meth-
- Page 1: FACIAL SOFT BIOMETRICSMETHODS, APPL
- Page 5: AcknowledgementsThis thesis would n
- Page 8: 6hair, skin and clothes. The propos
- Page 11 and 12: 97 Practical implementation of soft
- Page 13 and 14: 11Notations used in this workE : st
- Page 15 and 16: 13Chapter 1IntroductionTraditional
- Page 17: 15event of collision, which is of s
- Page 21 and 22: 19Chapter 2Soft biometrics: charact
- Page 23 and 24: 21is the fusion of soft biometrics
- Page 25 and 26: 23plied on low resolution grey scal
- Page 27 and 28: 25Chapter 3Bag of facial soft biome
- Page 29 and 30: 27In this setting we clearly assign
- Page 31 and 32: 29Table 3.1: SBSs with symmetric tr
- Page 33 and 34: 31corresponding to p(n,ρ). Towards
- Page 35 and 36: the same category (all subjects in
- Page 37 and 38: 3.5.2 Analysis of interference patt
- Page 39 and 40: an SBS by increasing ρ, then what
- Page 41 and 42: 39Table 3.4: Example for a heuristi
- Page 43 and 44: 41for a given randomly chosen authe
- Page 45 and 46: 43Chapter 4Search pruning in video
- Page 47 and 48: 45Figure 4.1: System overview.SBS m
- Page 49 and 50: 472.52rate of decay of P(τ)1.510.5
- Page 51 and 52: 49to be the probability that the al
- Page 53 and 54: 51The following lemma describes the
- Page 55 and 56: 534.5.1 Typical behavior: average g
- Page 57 and 58: 55n = 50 subjects, out of which we
- Page 59 and 60: 5710.950.9pruning Gain r(vt)0.850.8
- Page 61 and 62: 59for one person, for trait t, t =
- Page 63 and 64: 61Chapter 5Frontal-to-side person r
- Page 65 and 66: 63Figure 5.1: Frontal / gallery and
- Page 67 and 68: 6510.90.80.7Skin colorHair colorShi
16 1. INTRODUCTIONsystems, and describe how the employed categorization algorithms (eye color detector, glasses andmoustache detector) are applied on a characteristic database <strong>of</strong> 646 people. In the same Section wefurthermore provide simulations that reveal the variability and range <strong>of</strong> the pruning benefits <strong>of</strong>feredby different SBSs. In Section 4.7 we derive concise closed form expressions on the measures <strong>of</strong>pruning gain and goodput, provide simulations, as well as derive and simulate aspects relating tothe complexity costs <strong>of</strong> different s<strong>of</strong>t biometric systems <strong>of</strong> interest.Chapter 5 - Frontal-to-side person re-identificationTypically biometric face-recognition algorithms are developed, trained, tested and improvedunder the simplifying assumption <strong>of</strong> frontal-to-frontal person recognition. Such algorithms thoughare challenged when facing scenarios that deviate from the training setting, such as for examplein the presence <strong>of</strong> non-constant viewpoints, including the frontal-to-side scenario. Most personrecognition algorithms, whether holistic or based on facial features, only manage to optimally handlepose differences that are less than about15 degrees. As a result, a variation in the pose is <strong>of</strong>tena more dominant factor than a variation <strong>of</strong> subjects. This aspect <strong>of</strong> pose variation comes to the forein video surveillance, where a suspect may be pictured firstly frontal, whereas the correspondingtest images could be captured from the side, thus introducing a frontal-to-side recognition problem.Towards handling this problem, we employ multiple s<strong>of</strong>t biometrics related traits. One <strong>of</strong>our tasks here is to get some insight into the significance <strong>of</strong> these traits, specifically the significance<strong>of</strong> using hair, skin and clothes patches for frontal-to-side re-identification. We are workingon the color FERET dataset [Fer11] with frontal gallery images for training, and side (pr<strong>of</strong>ile)probe images for testing. Towards achieving re-identification, the proposed algorithm first analyzesthe color in Section 5.2.4.1 and furthermore texture in Section 5.2.4.2 <strong>of</strong> the three patches.Then we study the intensity correlations between patches in Section 5.2.4.3. This analysis is thenfollowed by the construction <strong>of</strong> a single, stronger classifier that combines the above measures inSection 5.2.5, to re-identify the person from his or her pr<strong>of</strong>ile.Deviating from the above security related applications, we consider then an application closerto entertainment, and specifically consider the application <strong>of</strong> s<strong>of</strong>t biometrics in analyzing andquantifying facial aesthetics.Chapter 6 - S<strong>of</strong>t biometrics for quantifying and predicting facial aestheticsWith millions <strong>of</strong> images appearing daily on Facebook, Picasa, Flickr, or on different socialand dating sites, photographs are <strong>of</strong>ten seen as the carrier <strong>of</strong> the first and deciding impression <strong>of</strong> aperson. At the same time though, human perception <strong>of</strong> facial aesthetics in images is a priori highlysubjective.We related among others s<strong>of</strong>t biometric traits with this subjective human perception. In theprovided study we quantify insight on how basic measures can be used to improve photographsfor CVs or for different social and dating websites. This helps create an objective view on subjectiveefforts by experts / journalists when retouching images. We use the gained objective viewto examine facial aesthetics in terms <strong>of</strong> aging, facial surgery and a comparison <strong>of</strong> average femalesrelatively to selected females known for their beauty. Specifically in Section 6.3 we introducethe employed database, as well as describe the basic features and methods used in this study. InSection 6.4 we proceed with numerical results, and provide intuition on the role <strong>of</strong> features, imagequality and facial features, in human perception. In Section 6.5, we use these accumulated conclusionsto construct a basic linear model that predicts attractiveness in facial photographs usingdifferent facial traits as well as image properties. We then examine and validate the designed met-