Kleinberg, Krista F. (2008) Facial anthropometry as an evidential tool in forensic image comparison. PhD thesis. http://theses.gla.ac.uk/245/ Copyright and moral rights for this thesis are retained by the author A copy can be downloaded for personal non-commercial research or study, without prior permission or charge This thesis cannot be reproduced or quoted extensively from without first obtaining permission in writing from the Author The content must not be changed in any way or sold commercially in any format or medium without the formal permission of the Author When referring to this work, full bibliographic details including the author, title, awarding institution and date of the thesis must be given Glasgow Theses Service http://theses.gla.ac.uk/ [email protected] FFaacciiaall AAnntthhrrooppoommeettrryy aass aann EEvviiddeennttiiaall TTooooll iinn FFoorreennssiicc IImmaaggee CCoommppaarriissoonn Submitted to the University of Glasgow for the degree of Doctor of Philosophy By Krista F. Kleinberg, RPh Forensic Medicine and Science Division of Cancer Sciences and Molecular Pathology March 2008 ©Krista F. Kleinberg Acknowledgements I would like to express my sincere appreciation to my supervisors, Dr. Robert A. Anderson, Dr. J Paul Siebert, Professor Peter Vanezis, and Professor A. Mike Burton, for their time, guidance, encouragement and patience in the development of my research project. I would like to give an extra special thank you to Dr. Anderson and Dr. Siebert. Their help during my time at the University of Glasgow cannot be over stated. The end result is not only my completed thesis but also my love of forensic science and the discovery of new ideas and thoughts. I wish to extend special thanks to my University Advisor, Dr. Mike Edward and colleagues Maria Vanezis and Sarah Sholl, and to Fran Ackermann who was instrumental in guiding me through the PhD process. I want also to extend my thanks to John C. Hay and to the volunteers who graciously lent their free time to help with the research process. For their continued help and support I wish to thank Elizabeth Stein and Yvonne Spence and lastly to my friends and family for their never-ending encouragement. To my Dad and Mom, Michael and Marcylyn, I owe my deepest gratitude for supporting my desire to return to school and to study an area that has meaning and importance in today’s world. Thank you so much for your love and support. ii List of Contents Acknowledgements ii List of Contents iii List of Tables vi List of Figures xii Glossary of Terms xv Summary 16 1 Forensic Science and Human Identification 19 1.1 Aims and Scope of Present Study 21 2 Craniofacial Identification 23 2.1 Introduction 23 2.2 Anthropometry 23 2.3 Photogrammetry 28 2.3.1 Introduction 28 2.3.2 Forensic Photogrammetry 29 2.4 Morphology 31 2.5 Superimposition 35 2.6 Facial Reconstruction 40 2.6.1 Traditional methods 40 2.6.2 3-D Facial Reconstruction Using Computer Graphics 41 2.7 Eye Witness Identification 44 2.7.1 Police Line-ups 44 2.7.2 Mistaken Identification 46 2.7.3 Psychological Facial Recognition 47 2.8 Identification through Biometrics 48 2.8.1 Security Biometric System: Facial Recognition 50 2.8.2 Pose Invariant Facial Recognition Systems 52 2.9 Video Surveillance 54 2.9.1 The Role of Video Surveillance in Crime Detection/Prevention 54 2.9.2 Positive and Negative Aspects of Video Surveillance 55 2.10 Forensic Image Analysis 57 2.10.1 Resolution 57 2.10.2 Digital Image Enhancement Guidelines 57 2.11 Police Photographs (“Mug Shots”) in the United Kingdom 59 iii 2.11.1 Trends from 1989 59 2.11.2 Forensic Identification National Database (FIND) 60 3 Facial Landmarks 62 3.1 Introduction 62 3.2 Landmarks 62 3.3 Linear Measurements 69 3.4 Proportions 71 4 Pilot Anthropometry Laboratory Study of Still Images 74 4.1 Introduction and Aims 74 4.2 Experimental Section 74 4.2.1 Subjects 74 4.2.2 Equipment and Photography 75 4.3 Derivation of Anthropometric Measurements 78 4.3.1 Data Processing 79 4.3.2 Tests Used 80 4.4 Results 80 4.5 Discussion 86 4.5.1 Experimental Set-up 86 4.5.2 Landmarks 87 4.5.3 Proportions and Angle Measurements 88 4.5.4 Modelling the Data 90 4.6 Conclusions 95 5 Investigation of Uncertainty of Anthropometric Measurements 97 5.1 Introduction and Aims 97 5.2 Errors in Facial Proportions Resulting from Variations in Landmark Placement 98 5.2.1 Materials and methodology 98 5.2.2 Results 100 5.3 Errors in Facial Proportions Resulting from the Process of Taking Photographs 103 5.3.1 Materials 103 5.3.2 Methodology 105 5.3.3 Results 109 5.4 Estimation of Uncertainty 121 5.5 Discussion 124 5.5.1 Introduction 124 iv 5.5.2 Landmark placement 124 6 Anthropometry Study of Paired Still and Video Images 130 6.1 Introduction 130 6.2 Materials and Methodology 130 6.2.1 Subjects 130 6.2.2 Data Collection 132 6.2.3 Experimental Design 132 6.3 Results 140 6.3.1 Comparisons within a Single Sample 140 6.3.2 Re-landmarked Images in Sample 1 145 6.3.3 Between Sample Comparisons 149 6.3.4 Removal of Lowest Variant Elements 165 6.3.5 Subsample of Images Requiring Human Verification 177 6.4 Discussion 178 6.4.1 Overview of the Study 179 6.4.2 Limitations of the Study and Further Work 186 6.4.3 Conclusions 188 7 Discussion of Anthropometry 190 7.1 Probability 191 7.2 Landmark Placement 193 7.3 Other Research into Anthropometric Analysis 195 7.4 Conclusions 198 8 Final Conclusions and Suggestions for Further Work 203 8.1 A Retrospective Overview of the Project 203 8.2 Future Research 205 8.3 Final Words 207 References 209 Appendix- Low Variant Proportions 224 Appendix- Morphological Analysis 226 Publications v List of Tables Table 2.1 Comparison of anthropometric measurements taken from a Bust of John Paul Jones and the corpse in question [23] ...Error! Bookmark not defined. Table 3.1 Definitions of Landmarks Used in this Study..............................63 Table 3.2 Landmarks Used in the Study by Purkait [148] ...........................66 Table 3.3 Linear Measurements used in the Present Study.........................70 Table 3.4 Proportions selected for the Present Study ...............................72 Table 4.1 List of five subjects and their nationalities.................................75 Tables 4.2a-b Means and Relative Standard Deviations for (a) Proportions and (b) Angles in five subjects ....................................................................81 Table 5.1 Average and range of coefficients of variation for measurements of proportions in video images and database photographs for intra-operator study. ..............................................................................................100 Table 5.2 Each image was measured 6 times by each operator. The table lists the average (n=6) and range of coefficients of variation obtained by each operator for measurements of proportions of each video image for inter-operator study......101 Table 5.3 The coefficient variation (CV %) of 59 proportions obtained from photographs of a subject taken by 10 operators at three different camera angles. Proportions with “ ’ ” indicate the left side of the face................................110 Table 5.4 The coefficient variation (CV %) of 59 proportions for the series of photographs in which a single volunteer took photographs of 10 different subjects. Proportions with “ ’ ” indicate the left side of the face................................114 Table 5.5 The coefficient variation (CV %) of 59 proportions for the series of photographs in which a single volunteer took photographs of a single subject. The height of the camera was not adjusted to accommodate the sloping floor. Proportions with “ ’ ” indicate the left side of the face................................116 Table 5.6 The coefficient variation (CV %) of 59 proportions for the series of photographs in which a single volunteer took photographs of a single subject. The height of the camera was adjusted to accommodate the sloping floor. Proportions with “ ’ ” indicate the left side of the face...............................................119 Table 5.7 Contributions to the uncertainty of anthropometric measurements ...122 Table 6.1 Definitions of two Samples used in the present research study.......132 Table 6.2 Definitions given to correct and incorrect face matches................134 Table 6.3 Relative frequency histograms illustrating comparisons of the mean absolute differences between proportions in true negative faces within a single sample......................................................................................141 Table 6.4 Relative frequency histograms illustrating comparisons of the Euclidean distances between proportions in true negative faces within a single sample....142 Table 6.5 Relative frequency histograms illustrating comparisons of Cosine θ distance equations (raw values) in true negative faces within a single sample for Sample 1 (a) and Sample 2 (b). Normal curves are superimposed in (c). ........144 Table 6.6 Relative frequency histograms illustrating comparisons of Cosine θ distance equations (mean subtracted) in true negative faces within a single sample for Sample 1 (a) and Sample 2 (b). Normal curves are superimposed in (c). ....144 Table 6.7 Relative frequency histograms illustrating comparisons of Cosine θ distance equations (Z-normalized) in true negative faces within a single sample for Sample 1 (a) and Sample 2 (b). Normal curves are superimposed in (c). ........145 Table 6.8 Superimposed normal curve histograms illustrating true positive and true negative face comparisons of the Cosine θ (raw and mean subtracted) distance equations in six re-landmarked images from Sample 1 using one experienced operator.....................................................................147 Table 6.9 Superimposed normal curve histograms illustrating true positive and true negative face comparisons of the Cosine θ (Z-normalized) distance equations in six re-landmarked images from Sample 1 using one experienced operator and all operators. ..............................................................................148 Table 6.10 Relative frequency histograms illustrating comparisons of the mean absolute difference between proportions of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d).......................................151 Table 6.11 Relative frequency histograms illustrating comparisons of the mean absolute differences between proportions of Sample 1 vs. the matched faces in Sample 2 using Z-normalized values showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d). ...........152 vii Table 6.12 Relative frequency histograms illustrating comparisons using the Euclidean distance between proportions of Sample 1 vs. the matched faces in Sample 2 using Z-normalized values showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d). ...........153 Table 6.13 Relative frequency histograms illustrating comparisons using the Cosine θ distance (raw) between proportions of Sample 1 vs. the matched faces in Sample 2 using Z-normalized values showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d). ...........154 Table 6.14 Relative frequency histograms illustrating comparisons using the log of proportional values to derive the Cosine θ (raw) distance of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d)....................156 Table 6.15 Relative frequency histograms illustrating comparisons using the log of proportional values to derive the Cosine θ (mean subtracted) distance of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d). ...........157 Table 6.16 Relative frequency histograms illustrating comparisons using the log of proportional values to derive the Cosine θ (Z-normalized) distance of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d). ...........158 Table 6.17 Relative frequency histograms illustrating comparisons using the Cosine θ (Z-normalized) distance of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d). Statistics to determine Z-normalized values were derived from Sample 1. ...........................................................160 Table 6.18 Relative frequency histograms illustrating comparisons using the Cosine θ (Z-normalized) distance of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d). Statistics to determine Z-normalized values were derived from Sample 2. ...........................................................161 Table 6.19 Relative frequency histograms illustrating comparisons using the Cosine θ (Z-normalized) distance of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), viii and both cases superimposed (d).Statistics to determine Z-normalized values were derived from a summation of the two samples.................................162 Table 6.20 Summary of the conditions imposed and results achieved in the comparison of faces in Sample 1 and Sample 2. Results are illustrated by the superimposed normal histogram curves showing the amount of overlap in true positive (dot) and true negative (solid) faces indicating the chance of misclassification...........................................................................163 Table 6.21 Relative frequency histograms of the element mean of Sample 1 and the element mean of Sample 2.......................................................166 Table 6.22 Relative frequency histograms of the element standard deviation for Sample 1 and the element standard deviation for Sample 2........................166 Table 6.23 Descending order of element standard deviations in each sample place on a line graph.....................................................................167 Table 6.24 Relative frequency histograms illustrating comparisons of Cosine θ distance equations (Z-normalized) within a single sample for Sample 1 (a) and Sample 2 (b) with 5% of lowest variant elements removed. Normal curves are superimposed in (c). .....................................................................169 Table 6.25 Relative frequency histograms illustrating comparisons of Cosine θ distance equations (Z-normalized) within a single sample for Sample 1 (a) and Sample 2 (b) with 10% of lowest variant elements removed. Normal curves are superimposed in (c). .....................................................................170 Table 6.26 Relative frequency histograms illustrating comparisons of Cosine θ distance equations (Z-normalized) within a single sample for Sample 1 (a) and Sample 2 (b) with 25% of lowest variant elements removed. Normal curves are superimposed in (c). .....................................................................170 Table 6.27 Relative frequency histograms illustrating comparisons of Cosine θ distance equations (Z-normalized) within a single sample for Sample 1 (a) and Sample 2 (b) with 50% of lowest variant elements removed. Normal curves are superimposed in (c). .....................................................................171 Table 6.28 Relative frequency histograms illustrating comparisons using the Cosine θ (Z-normalized) distance of Sample 1 vs. the matched faces in Sample 2 showing true positive cases (a), true negative cases (b), both cases together (c), and both cases superimposed (d) with 5% of lowest variant elements removed from vector.................................................................................172 ix