Crime scene reconstruction—Sex prediction from blood stained foot sole impressions Nabanita Basu, Samir Kumar Bandyopadhyay Forensic Science International Volume 278, Pages 156-172 (September 2017) DOI: 10.1016/j.forsciint.2017.06.017 Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 1 Data model developed for sex estimation from bloody broken crime scene footprint images. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 2 Right Foot sole outline of Volunteer 4 along with the 7 measurements used for footprint identification in forensic podiatry [Gunn’s Method] [5]. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 3 (a) Elongated Blood pool created with 10ml porcine blood on a plastic sheet. (b) Concentrated blood pool created with 10ml. blood on a plastic sheet. Differently shaped blood pools were created to establish that the patent footprint from the same individual can vary considerably based on the surface area of the lower side of the foot that has come in contact with blood. [In order to prevent coagulation of blood, blood collected from pig was processed with anticoagulant, Heparin]. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 4 Schematic representation of the footprints obtained from the male and female volunteers who consented to be a part of the experiment. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 5 Snapshot of the Footprint database that has been developed. The aspect ratio for each image was maintained when mounting the images on a canvas, so that each image in the database could be of the same dimension (250×400 pixel2). Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 6 Schematic representation of the image preprocessing technique/methodology that has been used for removing ruler artifacts, shadow effect and noise from the image. The technique also facilitates segmentation of the Region of Interest (ROI) from the raw image file. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 7 Stepwise schematic representation of the Image processing method output when applied to a heavily shadowed image. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 8 Ensemble algorithm for foot type (i.e. left/right foot) prediction based on the majority votes cast by the 3 rule based functions (namely, Function A, B and C). Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 9 The images represent the output for the 3 algorithms used for detecting whether the footprint impression was obtained from the left or the right foot of an individual irrespective of sex. Figure (a) represents the triangles drawn based on Function A while Figure (b) and (c) represent the triangles drawn or overlaid on the foot image based on Function B and C respectively. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 10 An image of the right foot of an individual representing the angle by which the image has been rotated. The image is rotated about the point at which the green and magenta lines intersect. The image is rotated anticlockwise if (90-Ø) is negative. If (90-Ø) is positive the image is rotated in clockwise direction. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 11 (a) The original image. (b) The rotated image. [The size of the rotated image for most cases was larger than the size of the cropped Binary image]. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 12 (a) Represents the length and breadth calculated on the registered image (Pink Line-Length; Green Line-Maximum Width) (b) Represents the calculated walking angle (i.e. the angle between the blue line and dotted vertical pink line) (c) The image represents the geometric rectangular shape that was used to represent the proportions of a foot impression. The yellow and orange lines represent the measurements used to describe a footprint impression. [The colored points/lines pertain to color illustration of the Figure. Please refer to the online version of the article for the colored illustration.] Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 13 The Euclidean distance between the pair of points (L, M), (L, P) and (L, N) was used to define the geometrical approximation of the registered ROI. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 14 Algorithm devised for approximate geometric shape fitting to obtain an overall idea of the dimension of the footprint impression recorded. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 15 An approximation of the heel size for incomplete foot impression. [The colored points/lines pertain to color illustration of the Figure. Please refer to the online version of the article for the colored illustration.] Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 16 (a) Represents the maximal accuracy levels achieved by a 1-NN classifier for wrapper feature sets of different size. The performance of the automated system for the wrapper feature sets selected for the manual system has also been represented in the graph. (b) Wrapper feature sets with maximum accuracy levels for the automated system were also used to evaluate the performance of the manual system. The graphs developed are comparable in either of the two cases across all wrapper feature set sizes. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 17 Represents the datapoints plotted in a 2D feature space defined by Feature 3 and 13 (Table 6). Features 3 and 13 have been selected for the display as because as a wrapper feature set they recorded the highest accuracy value for the Automated System (81.82%). This graph particularly emphasizes how the features (3 and 13) wrap around a 1-NN classifier and might eventually not provide satisfactory results for a 3-NN classifier. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 18 Represents the accuracy, sensitivity and specificity values for the test set across different Relief Feature Set size, against the accuracy values of the training set based on which the filter feature set was developed. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 19 (a) The ROC curve has been used as a performance metric to evaluate the performance of the Manual Wrapper and Automated Wrapper System against an Unbiased Random Sex Predictor System. The same metric has also been used to analyze the performance of an (b) Automated Filter Feature set based system and Manual System against an Unbiased Random System having an AUC value of 50 units. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 20 Bloodstained footprint impression on patterned surfaces. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions
Fig. 21 The output for the user assisted ROI extraction and subsequent automated feature extraction. Forensic Science International 2017 278, 156-172DOI: (10.1016/j.forsciint.2017.06.017) Copyright © 2017 Elsevier B.V. Terms and Conditions