Norm-based Coding Research Articles

Coding spatial variations in faces and simple shapes: a test of two models.

Faces all have the same basic elements in the same overall arrangement, and must be discriminated using variations in this shared configuration. An efficient way to represent these variations would be to code how each configuration differs from an average face (norm-based coding model). Alternatively, configurations could be represented simply by coding their absolute values in some coordinate system (absolute coding model). The two models differ in the variables they predict will influence an image's recognizability. Absolute coding predicts that recognizability will depend on an image's distinctiveness and degree of distortion from its veridical target. Norm-based coding predicts that recognizability will also depend on the way the image differs from a norm or average face, namely its distance from the norm and/or its degree of displacement from the norm-deviation vector for the target. We determined the effects of these four critical variables on recognition of undistorted (veridical) images, caricatures, anticaricatures and 'lateral' distortions of famous faces (Experiment 1), newly learned faces (Experiment 2), and simple shapes that also share a configuration (Experiment 2). The results favored absolute coding of both faces and shapes, and indicate that caricatures derive their power from their distinctiveness.

Categorization and Perceptual Learning: An Analogue of the Face Inversion Effect

This paper reports two experiments that investigate the extent to which it is plausible to suppose that an associatively based mechanism for perceptual learning acts as the basis for the effects of inversion on identification, recognition, matching and discrimination of faces (and certain other stimuli rendered familiar by expertise, e.g. gundogs). In the first experiment, an inversion effect that is contingent both on familiarity with a category and on the category possessing prototypical structure is demonstrated using discrimination learning of chequerboard stimuli. The second experiment demonstrates that the inversion effect found in Experiment 1 can generalize to a recognition paradigm as well. These results are discussed within the framework provided by associative learning theory, and a parallel is drawn with models employing a norm-based coding in similarity space. The conclusion is that it would be remarkable if the inversion effects demonstrated with the abstract categories used in the experiments reported here were not implicated in the inversion effects found with other classes of stimuli, whilst conceding that the analogy is not complete, particularly in the case of faces.

Understanding face recognition: Caricauture effects, inversion, and the homogeneity problem

Abstract Faces and other objects that share a configuration present a special problem to the visual system. Two components of the visual system's solution to this homogeneity problem have been identified. Inversion studies have identified the use of relational features (Diamond & Carey, 1986; Rhodes, Brake, & Atkinson, 1993), and caricature studies have identified norm-based coding (Carey, Rhodes, Diamond, & Hamilton, in preparation; Rhodes, Brennan, & Carey, 1987; Rhodes & McLean, 1990). Here we explore a possible link between these two components, asking whether caricature effects depend selectively on exaggeration of relational features. If so, then inversion, which makes relational features particularly difficult to code (compared with isolated features), should reduce caricature effects. In three experiments we found a caricature equivalence effect (caricatures identified as accurately as undistorted images and both better than anticaricatures) that was unaffected by orientation, suggesting that relational feature coding is not necessary for caricatures to be effective. Therefore, caricature and inversion effects reflect distinct components of face recognition. Caricature level and orientation also interacted differently with other factors, as would be expected if their effects depend upon different underlying processes. For one set of faces there was a caricature advantage in accuracy (Experiments 1 and 3). This superportrait effect occurred even for subjects who were unfamiliar with the faces prior to the experiment (Experiment 3), a result with important forensic implications. Furthermore, the effect was restricted to upright faces. Therefm, although caricatures can be recognized as well as undistorted images whatever features are exaggerated, exaggeration of relational features may be needed for a superportrait effect.

Visual processing of facial distinctiveness.

Three experiments are reported in which the role of facial distinctiveness in linedrawing representations of famous faces has been examined. In the first subjects were asked to select their impressions of best likeness interactively from a continuous range of drawings generated in real-time. On average, 42% of caricatures were considered to be the best likeness of famous individuals. The distinctiveness of the target faces related by independent observers correlated with the degree of exaggeration faces received to produce an optimal likeness. More distinctive faces required less caricaturing. In the second experiment caricature levels chosen as best likenesses were investigated. Caricatures of famous faces were recognised more quickly (by 36% of subjects) and more accurately (by 28%) than the true veridical drawings. Thus caricature enhancement can improve recognition in terms of both speed and accuracy. In the third experiment the role of distinctiveness and its interaction with external and internal facial features was investigated. The external (outer) hairline present in the previous experiment was found to provide a powerful frame of reference for the perceptual assessment of individuals' appearances; estimations of best likeness without external features present were, on average, veridical. It is concluded that accurate line drawings of even highly familiar faces are poor at conveying identity. Distinctiveness enhancement improves their likeness and their recognisability. Such caricatured drawings provide improved access to memories of famous faces, which lends strong support to models of human-face memory and processing based on norm-based coding. When required to decide upon the veracity of line drawings based only upon internal features subjects were able to make accurate judgments of veridicality. As well as validating the interaction approach to best-likeness judgments, this result further indicates the difficulty in making judgments of identification based on piecemeal and sparse configural information.

Towards an Exemplar Model of Face Processing: The Effects of Race and Distinctiveness

Valentine (1991a, 1991b) described a theoretical framework for face recognition in which faces are encoded as locations in a multidimensional space. It was argued that this approach could provide a unified account of the effects of distinctiveness, inversion, and race on face recognition. In this paper we evaluate the ability of this theoretical framework to account for the effects of distinctiveness and race in four experiments in which white British and Japanese faces served as stimuli and both white British and Japanese students acted as subjects. In a recognition memory experiment the expected "own-race bias" was observed as a Race of Subject x Race of Face interaction. Distinctive faces were recognized more accurately than typical faces, but the effect of distinctiveness did not interact with the race of face or the race of subject. Typical faces were classified faster than distinctive faces in a task in which intact faces had to be distinguished from jumbled faces, as found in earlier work, and the effect of distinctiveness did not interact with the race of face or race of subject. In contrast, a task in which subjects classified faces according to their race did show a greater effect of distinctiveness for own-race faces. The results are discussed in relation to the two specific models within the multidimensional space framework identified by Valentine (1991a): a purely exemplar-based model and a norm-based coding model. It is argued that these results are more easily accommodated in terms of a purely exemplar-based model. Some conceptual problems in applying the norm-based coding model to the effect of race are discussed.

A Unified Account of the Effects of Distinctiveness, Inversion, and Race in Face Recognition

A framework is outlined in which individual faces are assumed to be encoded as a point in a multidimensional space, defined by dimensions that serve to discriminate faces. It is proposed that such a framework can account for the effects of distinctiveness, inversion, and race on recognition of faces. Two specific models within this framework are identified: a norm-based coding model, in which faces are encoded as vectors from a population norm or prototype, and a purely exemplar-based model. Both models make similar predictions, albeit in different ways, concerning the interactions between the effects of distinctiveness, inversion and race. These predictions were supported in five experiments in which photographs of faces served as stimuli. The norm-based coding version and the exemplar-based version of the framework cannot be distinguished on the basis of the experiments reported, but it is argued that a multidimensional space provides a useful heuristic framework to investigate recognition of faces. Finally, the relationship between the specific models is considered and an implementation in terms of parallel distributed processing is briefly discussed.