FACIAL SOFT BIOMETRICS - Library of Ph.D. Theses | EURASIP

15event of collision, which is of significant character when employing SBSs for person identification.Furthermore we introduce the number of effective categories F which is later identified as animportant parameter related to collision, and is shown to directly affect the overall performance ofan SBS.Section 3.5.2 analyzes the statistical distribution and mean ofF and furthermore Section 3.5.3offers an insight regarding the bounds of the statistical behavior ofF over large populations. Thesebounds address the following practical question: if more funds are spent towards increasing thequality of an SBS, then what reliability gains do we expect to see? The answer and further intuitionon the above bounds are provided in Section 3.5.3.1, the proofs can be found in the Appendix 3.In Section 3.5.4 we examine the influence of algorithmic estimation errors and give an exampleon the overall performance of a realistic SBS. We improve the performance by a study of thedistribution between population in the overall categories, see Section 3.5.4.1. We then proceed inSection 3.5.4.2 to elaborate on the human compliant aspect of soft biometrics in re–identification,hereby specifically on the quantification of traits and on the human interaction view of an SBS.Chapter 4 - Search pruning in video surveillance systemsIn Chapter 4 we explore the application using soft-biometric related categorization-based pruningto narrow down a large search.In recent years we have experienced an increasing need to structure and organize an exponentiallyexpanding volume of images and videos. Crucial to this effort is the often computationallyexpensive task of algorithmic search for specific elements placed at unknown locations inside largedata sets. To limit computational cost, soft biometrics pruning can be used, to quickly eliminate aportion of the initial data, an action which is then followed by a more precise and complex searchwithin the smaller subset of the remaining data. Such pruning methods can substantially speed upthe search, at the risk though of missing the target due to classification errors, thus reducing theoverall reliability. We are interested in analyzing this speed vs. reliability tradeoff, and we focuson the realistic setting where the search is time-constrained and where, as we will see later on, theenvironment in which the search takes place is stochastic, dynamically changing, and can causesearch errors. In our setting a time constrained search seeks to identify a subject from a large setof individuals. In this scenario, a set of subjects can be pruned by means of categorization that isbased on different combinations of soft biometric traits such as facial color, shapes or measurements.We clarify that we limit our use of “pruning the search” to refer to the categorization andsubsequent elimination of soft biometric-based categories, within the context of a search withinlarge databases (Figure 4.1). In the context of this work, the elimination or filtering our of theemployed categories is based on the soft biometric characteristics of the subjects. The pruneddatabase can be subsequently processed by humans or by a biometric such as face recognition.Towards analyzing the pruning behavior of such SBSs, Chapter 4 introduces the concept ofpruning gain which describes, as a function of pruning reliability, the multiplicative reductionof the set size after pruning. For example a pruning gain of 2 implies that pruning managed tohalve the size of the original set. Section 4.5.1 provides average case analysis of the pruning gain,as a function of reliability, whereas Section 4.5 provides atypical-case analysis, offering insighton how often pruning fails to be sufficiently helpful. In the process we provide some intuitionthrough examples on topics such as, how the system gain-reliability performance suffers withincreasing confusability of categories, or on whether searching for a rare looking subject rendersthe search performance more sensitive to increases in confusability, than searching for commonlooking subjects.In Section 4.6.1 we take a more practical approach and present nine different soft biometric