Peter J. B. Hancock

From Pixels to People: A Model of Familiar Face Recognition

Research in face recognition has largely been divided between those projects concerned with front-end image processing and those projects concerned with memory for familiar people. These perceptual and cognitive programmes of research have proceeded in parallel, with only limited mutual influence. In this paper we present a model of human face recognition which combines both a perceptual and a cognitive component. The perceptual front-end is based on principal components analysis of face images, and the cognitive back-end is based on a simple interactive activation and competition architecture. We demonstrate that this model has a much wider predictive range than either perceptual or cognitive models alone, and we show that this type of combination is necessary in order to analyse some important effects in human face recognition. In sum, the model takes varying images of “known” faces and delivers information about these people.

Viewing it differently: Social scene perception in Williams syndrome and Autism

The genetic disorder Williams syndrome (WS) is associated with a propulsion towards social stimuli and interactions with people. In contrast, the neuro-developmental disorder autism is characterised by social withdrawal and lack of interest in socially relevant information. Using eye-tracking techniques we investigate how individuals with these two neuro-developmental disorders associated with distinct social characteristics view scenes containing people. The way individuals with these disorders view social stimuli may impact upon successful social interactions and communication. Whilst individuals with autism spend less time than is typical viewing people and faces in static pictures of social interactions, the opposite is apparent for those with WS whereby exaggerated fixations are prevalent towards the eyes. The results suggest more attention should be drawn towards understanding the implications of atypical social preferences in WS, in the same way that attention has been drawn to the social deficits associated with autism.

The principal components of natural images

A neural net was used to analyse samples of natural images and text. For the natural images, components resemble derivatives of Gaussian operators, similar to those found in visual cortex and inferred from psychophysics. While the results from natural images do not depend on scale, those from text images are highly scale dependent. Convolution of one of the text components with an original image shows that it is sensitive to inter-word gaps.

Face processing: Human perception and principal components analysis

Principal components analysis (PCA) of face images is here related to subjects’ performance on the same images. In two experiments subjects were shown a set of faces and asked to rate them for distinctiveness. They were subsequently shown a superset of faces and asked to identify those that had appeared originally. Replicating previous work, we found that hits and false positives (FPs) did not correlate: Those faces easy to identify as being “seen” were unrelated to those faces easy to reject as being “unseen.” PCA was performed on three data sets: (1) face images with eye position standardized, (2) face images morphed to a standard template to remove shape information, and (3) the shape information from faces only. Analyses based on PCA of shape-free faces gave high predictions of FPs, whereas shape information itself contributed only to hits. Furthermore, whereas FPs were generally predictable from components early in the PCA, hits appeared to be accounted for by later components. We conclude that shape and “texture” (the image-based information remaining after morphing) may be used separately by the human face processing system, and that PCA of images offers a useful tool for understanding this system.

Robust representations for face recognition: the power of averages.

We are able to recognise familiar faces easily across large variations in image quality, though our ability to match unfamiliar faces is strikingly poor. Here we ask how the representation of a face changes as we become familiar with it. We use a simple image-averaging technique to derive abstract representations of known faces. Using Principal Components Analysis, we show that computational systems based on these averages consistently outperform systems based on collections of instances. Furthermore, the quality of the average improves as more images are used to derive it. These simulations are carried out with famous faces, over which we had no control of superficial image characteristics. We then present data from three experiments demonstrating that image averaging can also improve recognition by human observers. Finally, we describe how PCA on image averages appears to preserve identity-specific face information, while eliminating non-diagnostic pictorial information. We therefore suggest that this is a good candidate for a robust face representation.

Do faces capture the attention of individuals with Williams syndrome or Autism? Evidence from tracking eye movements

The neuro-developmental disorders of Williams syndrome (WS) and autism can reveal key components of social cognition. Eye-tracking techniques were applied in two tasks exploring attention to pictures containing faces. Images were (i) scrambled pictures containing faces or (ii) pictures of scenes with embedded faces. Compared to individuals who were developing typically, participants with WS and autism showed atypicalities of gaze behaviour. Individuals with WS showed prolonged face gaze across tasks, relating to the typical WS social phenotype. Participants with autism exhibited reduced face gaze, linking to a lack of interest in socially relevant information. The findings are interpreted in terms of wider issues regarding socio-cognition and attention mechanisms.

The role of masculinity and distinctiveness in judgments of human male facial attractiveness

Masculinity and distinctiveness have been found to influence the attractiveness of human male faces. The relationship between masculinity and distinctiveness, however, has received little attention. In Expt 1, we examine how current averaging techniques and manipulated sexual dimorphism influence ratings of attractiveness, masculinity, and distinctiveness. In agreement with previous studies, composite faces were found to be more attractive than individual faces. Averaging resulted in increased ratings of attractiveness but decreased ratings of masculinity and distinctiveness. This supports both that attractiveness is related to averageness and findings showing a preference for feminine traits in male faces. When controlling for attractiveness, no significant relationship was found between masculinity and distinctiveness. Manipulating sexual dimorphism did not alter distinctiveness ratings, indicating that feminized and masculinized faces are equally distinctive. These results are suggestive that masculinity and distinctiveness are separable components in face perception. In Expt 2, we look to improve on previous studies utilizing composite faces by examining how averaging in texture-only or shape-only changes perceptions of attractiveness, masculinity, and distinctiveness. Averaging in both shape and texture were found to increase attractiveness independently, showing that the increased attractiveness of composites is due to the combined action of these two manipulations.

An Empirical Comparison of Selection Methods in Evolutionary Algorithms

Selection methods in Evolutionary Algorithms, including Genetic Algorithms, Evolution Strategies (ES) and Evolutionary Programming, (EP) are compared by observing the rate of convergence on three idealised problems. The first considers selection only, the second introduces mutation as a source of variation, the third also adds in evaluation noise. Fitness proportionate selection suffers from scaling problems: a number of techniques to reduce these are illustrated. The sampling errors caused by roulette wheel and tournament selection are demonstrated. The EP selection model is shown to be equivalent to an ES model in one form, and surprisingly similar to fitness proportionate selection in another. Generational models are shown to be remarkably immune to evaluation noise, models that retain parents much less so.

A forensically valid comparison of facial composite systems

An evaluation of E-FIT, PROfit, Sketch, Photofit and EvoFIT composite construction techniques was carried out in a “forensically friendly format”: composites of unfamiliar targets were constructed from memory following a 3–4-hour delay using a Cognitive Interview and experienced operators. The main dependent variable was spontaneous naming and overall performance was low (10% average naming rate). E-FITs were named better than all techniques except PROfit, though E-FIT was superior to PROfit when the target was more distinctive. E-FIT, PROfit and Sketch were similar overall in a composite sorting task, but Sketch emerged best for more average-looking targets. Photofit performed poorly, as did EvoFIT, an experimental system. Overall, facial distinctiveness was found to be an important factor for composite naming.

A statistical analysis of natural images matches psychophysically derived orientation tuning curves

A neural net method is used to extract principal components from real-world images. The initial components are a Gaussian followed by horizontal and vertical operators, starting with the first derivative and moving to successively higher orders. Two of the components are ‘bar-detectors’. Their measured orientation selectivity is similar to that suggested by Foster & Ward (Proc. R. Soc. Lond. B 243, 75 (1991)) to account for brief-exposure psychophysical data. In tests with noise images, the ratio of sensitivity between the two components is controlled by the degree of anisotropy in the image.