UNIVERSITY OF SOUTHAMPTON COMP6211W1 SEMESTER 2 EXAMINATIONS 2014/15 BIOMETRICS Duration 120 MINS (2 Hours) This paper contains 6 questions Answer THREE questions Each question carries 33 marks. Only University approved calculators may be used. A foreign language word to word® translation dictionary (paper version) is permitted provided it contains no notes, additions or annotations. 5 page examination paper Copyright 2015© University of Southampton Page 1 of 5 Q1 a) Discuss the motivation for feature extraction in image processing and give examples of features that could be useful in an image classification. [8 marks] b) Describe three different metrics that could be used to measure the distance of image examples in feature space. [9 marks] c) Describe the k-nearest neighbour rule by giving a pseudo code implementation [9 marks] d) Describe how you could build a face recognition system using ideas in (a), (b) and (c). [7 marks] Q2 a) Explain how shapes can be described and then classified via Cartesian moment descriptions given by ∑∑= x y qp pq yxIyxm ),( where all symbols have their usual meanings. [11 marks] b) Given that the centre of mass ),( yx is given by 00 10 m mx = , 00 01 m my = , show that the following relationships hold for centralised moments: i) 010 =µ ii) 00 2 10 2020 m m m −=µ [16 marks] c) Describe how 20µ can be used to discriminate between the image of a ring and the image of a disc. [6 marks] Q3 a) Describe what mean, median, and mode filters are and which one is a linear filter. [10 marks] b) Explain which one of the filters in Q3a performs the best in the process of background subtraction in videos containing Gait biometric and why? [9 marks] c) Describe what the wavelet transform is and explain what are advantages and disadvantages of wavelet transform over Fourier transform. Provide two examples in biometric where wavelet Copyright 2015© University of Southampton Page 2 of 5 transform is used and explain what task the wavelet transform performs in these examples. [14 marks] Q4 a) Describe the basic assumptions concerning a biometric including that it should be accessible, revealable and unique. [9 Marks] b) Given an image of a subject as shown in Figure 1, describe suitable approaches that could be used to recognise a subject by information derived from the region of the nose (perhaps including its shape). Describe a complete system which can first detect this region and then extract features that can be used for recognition purposes. [24 Marks] Figure 1 Face Image (A Moorhouse, AN Evans, GA Atkinson, The nose on your face may not be so plain: Using the nose as a biometric, Proc. ICDP 2009) TURN OVER Copyright 2015© University of Southampton Page 3 of 5 Q5 a) Discuss the likely performance advantages and disadvantages of distance measures, and why statistics might be included in their formulation. [9 Marks] b) The Mahalanobis distance measure is given by ( ) ( )μpμp −∑−= −1TMAHd where the covariance matrix is formed of elements which express the variance as ( )( )[ ]jjiiij μpμp −−Ε=∑ where Ε denotes the expected (average) value. Describe what is meant by each term in these expressions. Given two dimensional measurements of subjects which are in clusters 1 and 2 (with means µ1 and µ2, standard deviations σ1 and σ2). For the three cases in Table 1, discuss whether recognition will be possible, and the structure of the covariance matrix in each case. µ1 µ2 σ1 σ2 Case 1 [2,2] [4,4] 1 1 Case 2 [2,2] [4,4] 4 4 Case 3 [2,2] [2.5,2.5] 0.1 0.1 Table 1 [18 Marks] c) For each of Cases 1,2 and 3 describe the likely outcome of using a more conventional distance measure, such as Euclidean or Manhattan distance. [6 Marks] Copyright 2015© University of Southampton Page 4 of 5 Q6 a) Discuss how a walking subject can be recognised from a sequence of images [9 Marks] b) Describe two methods by which a feature description can be derived, giving equations where appropriate. Describe the advantages and disadvantages of your chosen feature description. [18 Marks] c) If the subject walks past a bright lamp, describe the effect on one of your feature vectors and on your recognition process. [6 Marks] END OF PAPER Copyright 2015© University of Southampton Page 5 of 5
欢迎咨询51作业君