The role of the forensic hair examiner is to determine whether a questioned hair recovered from a crime scene could or could not be from the same source as a known sample and therefore whether it should be included or excluded as probative evidence. Arguably, traditional hair microscopy is a largely subjective process that relies heavily on the training and experience of the examiner. The aim of this project was to investigate three objective analytical methods—two based on image analysis and one based on infrared spectroscopy—to produce a hair examination protocol that balances qualitative microscopic observations with quantitative measures. First, numerical colour measurements were investigated for allocating hair to one of six nominal categories and for distinguishing one participant’s hair from another’s within a subpopulation of similar coloured hair. Overall, between 69.3 and 76.2% correct classification to the categories was achieved with the RGB and CIE L*a*b* colour models returning the highest prediction accuracy, and CIE XYZ colour model returning generally poor results, particularly among the darker hair categories. In an effort to refine and improve these results, analyses were repeated that incorporated only a limited set of categories and predictor variables. Correct allocation increased slightly for the dark hair categories while no improvement was observed for the light hair categories. For distinguishing between individual hairs within a subpopulation of similarly coloured hair, aside from the red hair category, the discriminating power was considered to be too low for the method to be recommended as a routine tool in forensic hair examination. Second, a novel image analysis technique was evaluated that involved applying threshold operations to image montages, in order to compare pigmentation characteristics in three separate hair populations―Fair, Medium and Dark shaded hair. The average pixel area of each black on white object and the length of the major and minor axes, as well as calculated measurements such as density, the percentage of small, medium and large objects, and the percentage of two nominal configurations―‘streaks’ and ‘clumps’―were evaluated as potential variables. The novel technique did not support discrimination between the selected participants. The Medium sample population resulted in the lowest number of images correctly allocated, with only 32% prediction accuracy, the Fair sample population resulted in 54% prediction accuracy and the Dark sample population showed the highest correct allocation, with 62% prediction accuracy. No obvious relationships were observed between each of the populations in terms of the number of variables selected, the strongest predicting variable or the overall prediction accuracy. Finally, ATR–FTIR spectroscopy was assessed for identifying trace contaminants—specifically, hair treatments—on the hair surface. Seven product types were evaluated following dense and sparse application to individual hairs. Infrared absorption peaks were apparent for six product types with only one type showing no significant absorption. Discriminant analysis comprising 254 wavenumbers between 1632 and 652 (at a spacing of 3.857 cm-1) resulted in 100% accuracy for 60 reference spectra of the products evaluated, albeit only one brand per product type was included in this analysis. The strongest predictor variables were generally between 1300 and 1000 cm-1 corresponding to the CO absorbance bands for ethers and esters, and at 1450 cm-1 corresponding to the CH3 asymmetrical bend vibration. Variations in the spectra most likely due to molecular interactions (e.g. hydrogen bonding), were observed following dense application to the hair. Only 73% of those spectra were correctly classified by product type. Following sparse application to hair, trace contaminants were not observed on the majority of samples. On the few samples where traces were observed, spectra of the product type could not be clearly resolved. Difficulties associated with improving the discriminating power of hair examinations were identified two decades ago, including that considerable intra-individual variation exists and that microscopic hair features are difficult to assess objectively (Robertson,1982). Emerging technologies could assist future examinations with classifying―or potentially individualising―forensic hair evidence. However, successful quantification and discrimination of hair characteristics has not yet been achieved, despite attempts made in this research. Until there is a universally applicable technique that will mimic microscopic analysis, current evaluations made by an experienced examiner are the best option available.
|Date of Award||2012|
|Supervisor||Christopher Lennard (Supervisor), James Robertson (Supervisor) & Claude Roux (Supervisor)|
Digital imaging and image processing techniques for the comparison of human hair features
McLaren, C. J. (Author). 2012
Student thesis: Doctoral Thesis