The advent of nuclear DNA (nuDNA) analysis altered the way forensic biology was both practised and viewed by the forensic biologists, police, the legal system and the general public. The ability of nuDNA to individualise analysis of evidence and attach a statistical frequency ratio to the result, created an expectation that numerical objectivity should be part of all forensic analysis. There are few scientists who would disagree with both the need and desirability of objective measures of their results. Forensic hair examiners are no exception as indicated by numerous scientific publications specifically discussing means of objectively assessing hair and its characteristics. While mitochondrial DNA offers a partially objective measure of hair the result is destructive of the sample. A method that objectively supports the hair analysts' microscopic findings and is non destructive would be beneficial to forensic hair examination. This project attempted to develop an objective measure of hair analysis by using both traditional light microscopic comparative techniques combined with a high end digital imaging and image analysis capacity. Where objectivity equals an empirical set of numbers that can be manipulated for statistical significance, the comparative biological sciences such as histology, anthropology and forensic hair examination struggle. Forensic hair examiners have long acknowledged the difficulty, even inability, of assigning numerical values to the features that characterise one hair as being different from another. The human scalp hair is a "morphological" unit that is not readily split into component parts or even that these parts lend themselves to a number value. There have been at least nine separate studies which favourably compare the specificity of microscopic hair examinations. The challenge this study addressed was to appraise the use of numerical features in forensic hair examination, with particular emphasis on those features currently resisting numerical evaluation; specifically, colour and pigmentary characteristics. The techniques used were based on obtaining high quality digital images, and using the pixels inherent in the images to obtain numerical values of such features as colour and pigmentation. The project sample was taken from the telogen scalp hairs obtained from the hairbrushes of ten nominally brown haired Caucasians, both male and female. The focus was twofold: o Compare colour analysis of hair images from brown haired Caucasians within three standard, internationally recognized colour models, namely Red-Green-Blue (RGB) colour model; CIE XYZ Tristimulus (1931) colour model; and CIE L*a*b* (1976) colour model. o Using the same sets of digital images, undertake pattern recognition analysis both intra and inter individual hair samples. Discriminate analysis of the mean colour values collected for each of the inherent colour variables in the three colour models (red, green, blue; X,Y,Z and L*,a*,b*) indicated the RGB colour model gave the least separation of brown haired individuals; CIE XYZ and CIE L*a*b* separated several individuals for all their individual samples and several other individuals were mostly separated with only one of their own samples overlapping with another. Pattern analysis used a small area that represented the overall pigment patterning observed along the length of the hair shaft. This area was extracted from the digital image within V++ Digital Optics image analysis software. The extracted pattern piece was then compared with other sample images within the same hair and four other hairs from the same individual. Pattern extracts were also compared between person hair samples. The comparisons generated a set of numerical values based on the pixel number on the "x" axis of the whole image and the average difference between the extracted pattern image and the whole image. Analysis of this data resulted in log distributions when persons were matched with themselves. It was also possible to refer an unknown pattern extract to this distribution and based on probabilities, predict as to whether or not the unknown sample fell within any of the known sample's distribution.
|Date of Award||2007|
|Supervisor||Ian MCNAUGHT (Supervisor) & James Robertson (Supervisor)|