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
24 2. SOFT BIOMETRICS: CHARACTERISTICS, ADVANTAGES AND RELATED WORKfrom videos and exploited in a multiple camera video surveillance scenario in order to track thesubjects that cross the surveillance network. In [LLZ06] an approach for recognizing the gender,ethnicity and age with facial images is proposed. The approach incorporates Gabor filter, Adaboostlearning as well as support vector machine classifiers. A further hybrid classification basedon gender and ethnicity is considered in [GPW98] and [GW99]. The hybrid approach consistsof an ensemble of radial basis function networks and inductive decision trees. The authors showrobustness and good performance. A different approach for analysis in hybrid soft biometric systemsis provided in [SGN08] and [RN10], where semantic information (which corresponds to softbiometric classifiers) is manually extracted from a series of videos. Using the analysis of variancethe authors select a pool of traits which are considered the most representative. Those traits arethen used together with gait information. The authors demonstrate that the additional informationprovided by the semantic traits increases the performance of the people recognition system basedon gait. Those results are extended in [RNS11] and in [RN11]. The authors in [ACPR10] go onestep further and study the relation of human body measures, which allows for certain applicationsthe prediction of missing body measures. In [VFT + 09] the authors propose an approach for peoplesearch in surveillance data, characterized by three main elements: sensors, body parts, and theirattributes. The body parts and attributes are here closely related to soft biometrics.2.4 Domains of applicationSoft biometrics are either employed as uni modal systems, classifying a single trait classifiers,or in a combination with other systems. We differentiate following main domains of application.Fusion with classical biometric traits : SBSs are incorporated in multi modal biometricalsystems with the goal of increasing the overall reliability. Such an approach has been followed, in[JDN04b], where the benefits of soft biometrics in addition to fingerprint lead to an improvementof approximately 5% over the primary biometric system.Pruning the search : SBS were employed in previous works to pre filter large biometricdatabases with the aim of higher search efficiency. Scientific work on using soft biometricsfor pruning the search can be found in [KBN08, KBBN09], where a multitude of attributes,like age, gender, hair and skin color were used for classification of a face database, as well asin [GBDB97, New95] where the impact of pruning traits like age, gender and race was identifiedin enhancing the performance of regular biometric systems.A third application is the employment of a multi modal SBS with the goal of human identificationor human re-identification.Human (re-)identification : For human (re-)identification the soft biometric trait related limitationsof distinctiveness and permanence are overcome by combining multiple traits. The concept ofBag of Soft Biometrics(BoSB) is directly inspired from the idea of Bag of Words [Joa98, WPS06]and Bag of Features [LSP06] developed under the context of text mining and content based imageretrieval. For the BoSB the “items” of the bag are soft biometric signatures extracted from thevisual appearance of the subject.Other possible applications relate to the ability to match people based on their biometric-traitpreferences, acquiring statistical properties of biometric identifiers of groups, avatar modellingbased on the instantaneous facial characteristics (glasses, beard or different hair color), statisticalsampling of audiences, and many others.
25Chapter 3Bag of facial soft biometrics for humanidentificationThe concept of person identification based on soft biometrics originates in the way humansperform face recognition. Specifically human minds decompose and hierarchically structure complexproblems into fractions and those fractions into further sub-fractions, cf. [Ley96], [Sim96].Consequently face recognition performed by humans is the division of the face into parts, andsubsequent classification of those parts into sub-categories. Those sub-categories are associatedwith what refer to as soft biometrics and the key is that each individual can be categorized in termsof such characteristics, by humans or by image processing algorithms. Although features suchas hair, eye and skin color, facial hair and shape, or body height and weight, gait, clothing colorand human metrology are generally non distinctive, a cumulative combination of such featuresprovides an increasingly refined and explicit description of a human.3.1 Main parameters: authentication group, traits, trait-instances,and categoriesThe setting of interest corresponds to the general scenario where, out of a large population, anauthentication group is randomly extracted as a random set ofnpeople, out of which one person ispicked for identification (and is different from all the other members of the authentication group).We note that this general scenario is consistent with both, the case of person verification as well asof identification. A general soft-biometric system employs detection that relates toλsoft biometrictraits (hair color, skin color, etc), where each trait i, i = 1,2,...,λ, is subdivided into µ i traitinstances, i.e., each trait i can take one of µ i values. We henceforth denote as category to be anyλ-tuple of different trait-instances, and we let Φ = {φ i } ρ i=1define a set of all ρ categories, i.e.,the set of all ρ combinations of soft-biometric trait-instances. The number of ρ, that the system isendowed with, is given byρ = Π λ i=1µ i (3.1)We slightly abuse notation and henceforth say that a subject belongs to category φ if hisor her trait-instances are the λ-tuple corresponding to category φ. We here note that to haveconclusive identification of a subject, and subsequent differentiation from the other subjects of theauthentication group, it must be the case that the subject does not belong in the same category asother members of the authentication group. Given a specific authentication group, the maximumlikelihoodoptimizing rule for detecting the most probable category in which a chosen subject
- 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 and 18: 15event of collision, which is of s
- Page 19 and 20: 17ric. In Section 6.6 we employ the
- Page 21 and 22: 19Chapter 2Soft biometrics: charact
- Page 23 and 24: 21is the fusion of soft biometrics
- Page 25: 23plied on low resolution grey scal
- 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
- Page 69 and 70: 6710.90.80.70.6Perr0.50.40.30.20.10
- Page 71 and 72: 69Chapter 6Soft biometrics for quan
- Page 73 and 74: 71raphy considerations include [BSS
- Page 75 and 76: 73Figure 6.3: Example image of the
24 2. <strong>SOFT</strong> <strong>BIOMETRICS</strong>: CHARACTERISTICS, ADVANTAGES AND RELATED WORKfrom videos and exploited in a multiple camera video surveillance scenario in order to track thesubjects that cross the surveillance network. In [LLZ06] an approach for recognizing the gender,ethnicity and age with facial images is proposed. The approach incorporates Gabor filter, Adaboostlearning as well as support vector machine classifiers. A further hybrid classification basedon gender and ethnicity is considered in [GPW98] and [GW99]. The hybrid approach consists<strong>of</strong> an ensemble <strong>of</strong> radial basis function networks and inductive decision trees. The authors showrobustness and good performance. A different approach for analysis in hybrid s<strong>of</strong>t biometric systemsis provided in [SGN08] and [RN10], where semantic information (which corresponds to s<strong>of</strong>tbiometric classifiers) is manually extracted from a series <strong>of</strong> videos. Using the analysis <strong>of</strong> variancethe authors select a pool <strong>of</strong> traits which are considered the most representative. Those traits arethen used together with gait information. The authors demonstrate that the additional informationprovided by the semantic traits increases the performance <strong>of</strong> the people recognition system basedon gait. Those results are extended in [RNS11] and in [RN11]. The authors in [ACPR10] go onestep further and study the relation <strong>of</strong> human body measures, which allows for certain applicationsthe prediction <strong>of</strong> missing body measures. In [VFT + 09] the authors propose an approach for peoplesearch in surveillance data, characterized by three main elements: sensors, body parts, and theirattributes. The body parts and attributes are here closely related to s<strong>of</strong>t biometrics.2.4 Domains <strong>of</strong> applicationS<strong>of</strong>t biometrics are either employed as uni modal systems, classifying a single trait classifiers,or in a combination with other systems. We differentiate following main domains <strong>of</strong> application.Fusion with classical biometric traits : SBSs are incorporated in multi modal biometricalsystems with the goal <strong>of</strong> increasing the overall reliability. Such an approach has been followed, in[JDN04b], where the benefits <strong>of</strong> s<strong>of</strong>t biometrics in addition to fingerprint lead to an improvement<strong>of</strong> approximately 5% over the primary biometric system.Pruning the search : SBS were employed in previous works to pre filter large biometricdatabases with the aim <strong>of</strong> higher search efficiency. Scientific work on using s<strong>of</strong>t biometricsfor pruning the search can be found in [KBN08, KBBN09], where a multitude <strong>of</strong> attributes,like age, gender, hair and skin color were used for classification <strong>of</strong> a face database, as well asin [GBDB97, New95] where the impact <strong>of</strong> pruning traits like age, gender and race was identifiedin enhancing the performance <strong>of</strong> regular biometric systems.A third application is the employment <strong>of</strong> a multi modal SBS with the goal <strong>of</strong> human identificationor human re-identification.Human (re-)identification : For human (re-)identification the s<strong>of</strong>t biometric trait related limitations<strong>of</strong> distinctiveness and permanence are overcome by combining multiple traits. The concept <strong>of</strong>Bag <strong>of</strong> S<strong>of</strong>t Biometrics(BoSB) is directly inspired from the idea <strong>of</strong> Bag <strong>of</strong> Words [Joa98, WPS06]and Bag <strong>of</strong> Features [LSP06] developed under the context <strong>of</strong> text mining and content based imageretrieval. For the BoSB the “items” <strong>of</strong> the bag are s<strong>of</strong>t biometric signatures extracted from thevisual appearance <strong>of</strong> the subject.Other possible applications relate to the ability to match people based on their biometric-traitpreferences, acquiring statistical properties <strong>of</strong> biometric identifiers <strong>of</strong> groups, avatar modellingbased on the instantaneous facial characteristics (glasses, beard or different hair color), statisticalsampling <strong>of</strong> audiences, and many others.