Benjamin Balas

Personal familiarity influences the processing of upright and inverted faces in infants

Infant face processing becomes more selective during the first year of life as a function of varying experience with distinct face categories defined by species, race, and age. Given that any individual face belongs to many such categories (e.g. A young Caucasian mans face) we asked how the neural selectivity for one aspect of facial appearance was affected by category membership along another dimension of variability. 6-month-old infants were shown upright and inverted pictures of either their own mother or a stranger while event-related potentials (ERPs) were recorded. We found that the amplitude of the P400 (a face-sensitive ERP component) was only sensitive to the orientation of the mothers face, suggesting that “tuning” of the neural response to faces is realized jointly across multiple dimensions of face appearance.

Face Recognition by Humans: Nineteen Results All Computer Vision Researchers Should Know About

A key goal of computer vision researchers is to create automated face recognition systems that can equal, and eventually surpass, human performance. To this end, it is imperative that computational researchers know of the key findings from experimental studies of face recognition by humans. These findings provide insights into the nature of cues that the human visual system relies upon for achieving its impressive performance and serve as the building blocks for efforts to artificially emulate these abilities. In this paper, we present what we believe are 19 basic results, with implications for the design of computational systems. Each result is described briefly and appropriate pointers are provided to permit an in-depth study of any particular result

Texture synthesis and perception: Using computational models to study texture representations in the human visual system

Traditionally, texture perception has been studied using artificial textures made of random dots or repeated shapes. At the same time, computer algorithms for natural texture synthesis have improved dramatically. We seek to unify these two fields through a psychophysical assessment of a particular computational model, providing insight into which statistics are most vital for natural texture perception. We employ Portilla and Simoncellis texture synthesis algorithm, a parametric model that mimics computations carried out in human vision. We find an intriguing interaction between texture type (periodic, structured, or 3-D textures) and image statistics (autocorrelation function and filter magnitude correlations), suggesting different representations may be employed for these texture families under pre-attentive viewing.

The Effect of Real-World Personal Familiarity on the Speed of Face Information Processing

Background Previous studies have explored the effects of familiarity on various kinds of visual face judgments, yet the role of familiarity in face processing is not fully understood. Across different face judgments and stimulus sets, the data is equivocal as to whether or not familiarity impacts recognition processes. Methodology/Principal Findings Here, we examine the effect of real-world personal familiarity in three simple delayed-match-to-sample tasks in which subjects were required to match faces on the basis of orientation (upright v. inverted), gender and identity. We find that subjects had a significant speed advantage with familiar faces in all three tasks, with large effects for the gender and identity matching tasks. Conclusion/Significance Our data indicates that real-world experience with a face exerts a powerful influence on face processing in tasks where identity information is irrelevant, even in tasks that could in principle be solved via low-level cues. These results underscore the importance of experience in shaping visual recognition processes.

Region-based representations for face recognition

Face recognition is one of the most important applied aspects of visual perception. To create an automated face-recognition system, the fundamental challenge is that of finding useful features. In this paper, we suggest a new class of image features that may be a useful addition to the set of representational tools for face-recognition tasks. Our proposal is motivated by the observation that rather than relying exclusively on traditional edge-based image representations, it may be useful to also employ region-based strategies that can compare noncontiguous image regions. The spatial homogeneity within regions allows for enhanced tolerance to geometric distortions and greater freedom in the choice of sample points. We first show that under certain circumstances, comparisons between spatially disjoint image regions are, on average, more valuable for recognition than features that measure local contrast. Second, we learn “optimal” sets of region comparisons for recognizing faces across varying pose and illumination. We propose a representational primitive---the dissociated dipole---that permits an integration of edge-based and region-based representations. This primitive is then evaluated using the FERET database of face images and then compared to established local and global algorithms.

N170 face specificity and face memory depend on hometown size.

Face recognition depends on visual experience in a number of different ways. Infrequent exposure to faces belonging to categories defined by species, age, or race can lead to diminished memory for and discrimination between members of those categories relative to faces belonging to categories that dominate an observers environment. Early visual impairment can also have long-lasting and broad effects on face discrimination - just a few months of visual impairment due to congenital cataracts can lead to diminished discrimination between faces that differ in their configuration, for example (Le Grand et al., 2001). Presently, we consider a novel aspect of visual experience that may impact face recognition: The approximate amount of different faces observers encountered during their childhood. We recruited undergraduate observers from small (500-1000 individuals) and large communities (30,000-100,000 individuals) and asked them to complete a standard face memory test and a basic ERP paradigm designed to elicit a robust N170 response, including the classic face inversion effect. We predicted that growing up in a small community might lead to diminished face memory and an N170 response that was less specific to faces. These predictions were confirmed, suggesting that the sheer number of faces one can interact with during their upbringing shapes their behavioral abilities and the functional architecture of face processing in the brain.

You can take the eyes out of the doll, but....

The perceived animacy of a face is well-predicted by the perceived animacy of the eyes presented in isolation. This is not true for other facial features (eg having a highly life-like nose does not appear to be crucial), suggesting that the eyes are a critical feature for perceiving life in a face. Here, we asked whether it was therefore possible to ‘transplant’ animacy into a face by transplanting the eyes into a face image. We conducted digital eye surgery on a series of morphed human/doll faces and found that while doll eyes make a morphed face look less alive, human eyes do not make you look more so. Thus, we cannot so easily transplant animacy into a face, but we can take it away.

Developing a Side Bias for Conspecific Faces during Childhood.

Adults preferentially use information from the left side of face images to judge gender, emotion, and identity. In this study, we examined the development of this visual-field bias over middle childhood (5-10 years). Our goal was to both characterize the developmental trajectory of the left-side bias (should one exist) and examine the selectivity of the phenomenon. We used own- versus other-species faces (human and monkey faces) to ask whether the left-side bias was equally strong for categories with which children have vastly different amounts of experience. We found that the left-side bias did show both a developmental trend over the age range we considered and distinct category selectivity; for human faces the preference for the left side of the image increased across the age range tested, but for monkey faces it did not. We discuss our results in the context of experience-dependent perceptual narrowing during development.

A speed-dependent inversion effect in dynamic object matching

The representations employed by the visual system for dynamic object recognition remain relatively unclear, due in part to the lack of sufficient data constraining the nature of the underlying encoding processes. In the current study, we examined the limits of invariant recognition for unfamiliar moving objects in the context of a same/different matching task. In Experiments 1 and 2, Observers were asked to evaluate whether pairs of moving objects differed in identity subject to a spatial manipulation (inversion) and a spatiotemporal manipulation (speed change between sample and target). We find evidence of a speed-dependent inversion effect, suggesting distinct modes of processing for fast-moving and slow-moving objects. Furthermore, we observe a deleterious effect of speed change between sample and test stimuli, indicating that the speed of appearance change is encoded by the visual system for recognition. In a third experiment, we also observed a speed-dependency in the extent to which the direction of motion is encoded by the visual system for recognition. These results are discussed in the context of previous proposals regarding dynamic object representations, and in terms of an emerging model of dynamic object perception.

Observinga Object Motion Induces Increased Generalization and Sensitivity

Learning to recognize a new object requires binding together dissimilar images of that object into a common representation. Temporal proximity is a useful computational cue for learning invariant representations. We report experiments that demonstrate two distinct psychophysical effects of temporal association via observed object motion on object perception. First, we use an implicit priming criterion to demonstrate that observation of a dynamic object induces generalization over close temporal neighbors. Second, in contrast to predictions from previous work, we find that shape discrimination between images actually improves following the same training procedure. We suggest that these apparently conflicting sets of results, one demonstrating blurring and the other demonstrating sharpening of the perceived distinction between temporally proximate frames, are consistent with a highly redundant code for object appearance.