Nigel D Haig

The Effect of Feature Displacement on Face Recognition

Human beings possess a remarkable ability to recognise familiar faces quickly and without apparent effort. In spite of this facility, the mechanisms of visual recognition remain tantalisingly obscure. An experiment is reported in which image processing equipment was used to displace slightly the features of a set of original facial images to form groups of modified images. Observers were then required to indicate whether they were being shown the “original” or a “modified” face, when shown one face at a time on a TV monitor screen. Memory reinforcement was provided by displaying the original face at another screen position, between presentations. The data show, inter alia, the very high significance of the vertical positioning of the mouth, followed by eyes, and then the nose, as well as high sensitivity to close-set eyes, coupled with marked insensitivity to wide-set eyes. Implications of the results for the use of recognition aids such as Identikit and Photofit are briefly discussed.

Exploring Recognition with Interchanged Facial Features

Any attempt to unravel the mechanism underlying the process of human face recognition must begin with experiments that explore human sensitivity to differences between a perceived image and an original memory trace. A set of three consecutive experiments are reported that were collectively designed to measure the relative importance of different facial features. The method involved the use of image-processing equipment to interchange cardinal features among frontally viewed target faces. Observers were required to indicate which of the original target faces most resembled the modified faces. The results clearly establish the dominant influence of the head outline as the major recognition feature. Next in importance is the eye/eyebrow combination, followed by the mouth, and then the nose. As a recognition feature in a frontally presented face, the nose is hardly noticed. The number of apparently random responses to some faces indicates that a surprisingly different face can sometimes arise from a fortuitous combination of the old features.

The Effect of Feature Displacement on the Perception of Well-Known Faces

The effect of feature displacement within two well-known faces (Terry Wogan and Cyril Smith) was examined. Image processing equipment was used to produce stimuli in which the features of an original facial image were displaced to form a number of modified images. This technique was first reported by Haig, in a recognition study in which the effect of feature displacement within unfamiliar faces was investigated. In the present experiment a perceptual judgement task was carried out in which subjects were presented with a number of modified faces and asked to judge how dissimilar these were with respect to an original image. A multidimensional scaling analysis of the comparative judgements of the subjects revealed a two-dimensional solution involving displacement of the eyes and mouth. A clear division between up/down and inward/outward displacement within these features (particularly the eyes) was observed. A similar pattern of results was found for both well-known faces. This result indicates that subjects were responding to changes in the facial configuration produced by the different types of feature displacement (horizontal or vertical), as opposed to movement of the features per se. Finally, the results also indicate that the displacement of inner features (mouth, eyes, nose) was more noticeable than displacement of the outer facial features (eg hairline).

How Faces Differ—A New Comparative Technique

It can be argued that the process of recognising faces progresses in two stages: first, the realisation that a perceived image contains patterns that may most reasonably be interpreted as forming a discrete face; second, correct and positive identification by noting the particular features that differentiate one face from all others. A novel technique which explored the latter process in the particular case of four different (male) faces is described. The experiment took the form of a four-alternatives forced-choice presentation of faces behind masks which contained a number of randomly positioned apertures. The percentage of correct responses for each separate aperture was then computed after a large number of 1 s presentations to four observers. This novel form of experiment suggested an equally novel form of pictorial data presentation that, literally, highlights the salient features of each individual face and thereby allows detailed intercomparison merely by inspection. Summing over all targets and observers reveals a strong preference for eyes and eyebrows, followed closely by the hairline above the temples. Next in order of preference comes the mouth and upper-lip area, followed by the lateral hairline beside each temple. Individual differences are strong, however, and the variations are such as to suggest that the uncritical application of generalised feature saliency lists is neither useful nor appropriate.

An investigation of trained neural networks from a neurophysiological perspective

The application of theoretical neural networks to preprocessed images was investigated with the aim of developing a computational recognition system. The neural networks were trained by means of a back-propagation algorithm, to respond selectively to computer-generated bars and edges. The receptive fields of the trained networks were then mapped, in terms of both their synaptic weights and their responses to spot stimuli. There was a direct relationship between the pattern of weights on the inputs to the hidden units (the units in the intermediate layer between the input and the output units), and their receptive field as mapped by spot stimuli. This relationship was not sustained at the level of the output units in that their spot-mapped responses failed to correspond either with the weights of the connections from the hidden units to the output units, or with a qualitative analysis of the networks. Part of this discrepancy may be ascribed to the output function used in the back-propagation algorithm.

A self-similar stack model for human and machine vision

A new model is proposed that not only exhibits the major properties of primate spatial vision but also has a structure that can be implemented efficiently in a machine vision system. The model is based on a self-similar stack structure with a spatial resolution that varies with eccentricity. It correctly reproduces the visual cortical mapping function, yet it has the important attribute that it can produce invariant responses to local changes in the size and position of image features. By proposing a novel purpose for cortical “bar-detectors”, the model can also produce invariance to more general distortions. The structure of the model allows efficient hierarchical search to be made and it naturally embraces the concept of “attention area”. Exploitation of this model has already confirmed these properties and has also revealed its robust ability to control the focus and gain of machine vision systems.

High-Resolution Facial Feature Saliency Mapping

For recognition of a target there must be some form of comparison process between the image of that target and a stored representation of that target. In the case of faces there must be a very large number of such stored representations, yet human beings seem able to perform comparisons at phenomenal speed. It is possible that faces are memorised by fitting unusual features or combinations of features onto a bland prototypical face, and such a data-compression technique would help to explain our computational speed. If humans do indeed function in this fashion, it is necessary to ask just what are the features that distinguish one face from another, and also, what are the features that form the basic set of the prototypical face. The distributed apertures technique was further developed in an attempt to answer both questions. Four target faces, stored in an image-processing computer, were each divided up into 162 contiguous squares that could be displayed in their correct positions in any combination of 24 or fewer squares. Each observer was required to judge which of the four target faces was displayed during a 1 s presentation, and the proportion of correct responses for each individual square was computed. The resultant response distributions, displayed as brightness maps, give a vivid impression of the relative saliency of each feature square, both for the individual targets and for all of them combined. The results, while broadly confirming previous work, contain some very interesting and surprising details about the differences between the target faces.

The refractive index correction in luminescence spectroscopy

Abstract It is shown that the so-called refractive index or “ n 2 ” correction used when calculating luminescence quantum efficiencies is generally inappropriate. For accurate evaluations no refractive index correction is required provided that a simple modification to luminescence spectrometers is carried out.

Effects of wavefront aberration on visual instrument performance, and a consequential test technique

Assessment of the visual image quality produced by afocal optical systems is complicated, by both the coherent coupling between eye and instrument and the general lack of correlation between the modulation transfer function and visual performance. This paper summarizes previous results obtained by the authors on simulating primary wavefront aberrations and extends the results to more typical higher-order aberrations, using the same image processor as before. The results strongly support the adoption of a form of Strehl ratio measurement as an outstandingly simple, yet visually appropriate assessment method for afocal systems. A possible test layout is suggested.

A New Visual Illusion, and its Mechanism

A new visual illusion is reported, in which a sine-wave grating appears to tilt when doubly sheared perpendicularly to the grating lines. It is shown that the illusory percept is related to the Münsterberg and Café Wall illusions. The probable mechanism at the root of all such illusions is postulated by reference to the neuroarchitecture of the retina and striate cortex.