Edges and bars: where do people see features in 1-D images?

There have been two main approaches to feature detection in human and computer vision - based either on the luminance distribution and its spatial derivatives, or on the spatial distribution of local contrast energy. Thus, bars and edges might arise from peaks of luminance and luminance gradient respectively, or bars and edges might be found at peaks of local energy, where local phases are aligned across spatial frequency. This basic issue of definition is important because it guides more detailed models and interpretations of early vision. Which approach better describes the perceived positions of features in images? We used the class of 1-D images defined by Morrone and Burr in which the amplitude spectrum is that of a (partially blurred) square-wave and all Fourier components have a common phase. Observers used a cursor to mark where bars and edges were seen for different test phases (Experiment 1) or judged the spatial alignment of contours that had different phases (e.g. 0 degrees and 45 degrees ; Experiment 2). The feature positions defined by both tasks shifted systematically to the left or right according to the sign of the phase offset, increasing with the degree of blur. These shifts were well predicted by the location of luminance peaks (bars) and gradient peaks (edges), but not by energy peaks which (by design) predicted no shift at all. These results encourage models based on a Gaussian-derivative framework, but do not support the idea that human vision uses points of phase alignment to find local, first-order features. Nevertheless, we argue that both approaches are presently incomplete and a better understanding of early vision may combine insights from both. (C)2004 Elsevier Ltd. All rights reserved.

Contrast discrimination and pattern masking: contrast gain control with fixed additive noise

We studied the visual mechanisms that serve to encode spatial contrast at threshold and supra-threshold levels. In a 2AFC contrast-discrimination task, observers had to detect the presence of a vertical 1 cycle deg-1 test grating (of contrast dc) that was superimposed on a similar vertical 1 cycle deg-1 pedestal grating, whereas in pattern masking the test grating was accompanied by a very different masking grating (horizontal 1 cycle deg-1, or oblique 3 cycles deg-1). When expressed as threshold contrast (dc at 75% correct) versus mask contrast (c) our results confirm previous ones in showing a characteristic 'dipper function' for contrast discrimination but a smoothly increasing threshold for pattern masking. However, fresh insight is gained by analysing and modelling performance (p; percent correct) as a joint function of (c, dc) - the performance surface. In contrast discrimination, psychometric functions (p versus logdc) are markedly less steep when c is above threshold, but in pattern masking this reduction of slope did not occur. We explored a standard gain-control model with six free parameters. Three parameters control the contrast response of the detection mechanism and one parameter weights the mask contrast in the cross-channel suppression effect. We assume that signal-detection performance (d') is limited by additive noise of constant variance. Noise level and lapse rate are also fitted parameters of the model. We show that this model accounts very accurately for the whole performance surface in both types of masking, and thus explains the threshold functions and the pattern of variation in psychometric slopes. The cross-channel weight is about 0.20. The model shows that the mechanism response to contrast increment (dc) is linearised by the presence of pedestal contrasts but remains nonlinear in pattern masking.

The interaction of luminance and texture amplitude in depth perception

Previous studies have suggested separate channels for the detection of first-order luminance (LM) and second-order modulations of the local amplitude (AM) of a texture (Schofield and Georgeson, 1999 Vision Research 39 2697 - 2716; Georgeson and Schofield, 2002 Spatial Vision 16 59). It has also been shown that LM and AM mixtures with different phase relationships are easily separated in identification tasks, and (informally) appear very different with the in-phase compound (LM + AM), producing the most realistic depth percept. We investigated the role of these LM and AM components in depth perception. Stimuli consisted of a noise texture background with thin bars formed as local increments or decrements in luminance and/or noise amplitude. These stimuli appear as embossed surfaces with wide and narrow regions. When luminance and amplitude changes have the same sign and magnitude (LM + AM) the overall modulation is consistent with multiplicative shading, but this is not so when the two modulations have opposite sign (LM - AM). Keeping the AM modulation depth fixed at a suprathreshold level, we determined the amount of luminance contrast required for observers to correctly indicate the width (narrow or wide) of raised regions in the display. Performance (compared to the LM-only case) was facilitated by the presence of AM, but, unexpectedly, performance for LM - AM was even better than for LM + AM. Further tests suggested that this improvement in performance is not due to an increase in the detectability of luminance in the compound stimuli. Thus, contrary to previous findings, these results suggest the possibility of interaction between first-order and second-order mechanisms in depth perception.

Bars and edges: what makes a feature for human vision?

There have been two main approaches to feature detection in human and computer vision - luminance-based and energy-based. Bars and edges might arise from peaks of luminance and luminance gradient respectively, or bars and edges might be found at peaks of local energy, where local phases are aligned across spatial frequency. This basic issue of definition is important because it guides more detailed models and interpretations of early vision. Which approach better describes the perceived positions of elements in a 3-element contour-alignment task? We used the class of 1-D images defined by Morrone and Burr in which the amplitude spectrum is that of a (partially blurred) square wave and Fourier components in a given image have a common phase. Observers judged whether the centre element (eg ±458 phase) was to the left or right of the flanking pair (eg 0º phase). Lateral offset of the centre element was varied to find the point of subjective alignment from the fitted psychometric function. This point shifted systematically to the left or right according to the sign of the centre phase, increasing with the degree of blur. These shifts were well predicted by the location of luminance peaks and other derivative-based features, but not by energy peaks which (by design) predicted no shift at all. These results on contour alignment agree well with earlier ones from a more explicit feature-marking task, and strongly suggest that human vision does not use local energy peaks to locate basic first-order features. [Supported by the Wellcome Trust (ref: 056093)]

Neural adjustments to image blur

Blur is an intrinsic feature of retina images that varies widely across images and observers, yet the world still typically appears 'in focus'. Here we examine the putative role of neural adaptation1 in the human perception of image focus by measuring how blur judgments depended on the state of adaptation. Exposure to unfocused images has previously been shown to influence acuity and contrast sensitivity and here we show that adaptation can also profoundly affect the actual perception of image focus.

Seeing edge blur: receptive fields as multiscale neural templates

Edge blur is an important perceptual cue, but how does the visual system encode the degree of blur at edges? Blur could be measured by the width of the luminance gradient profile, peak ^ trough separation in the 2nd derivative profile, or the ratio of 1st-to-3rd derivative magnitudes. In template models, the system would store a set of templates of different sizes and find which one best fits the `signature' of the edge. The signature could be the luminance profile itself, or one of its spatial derivatives. I tested these possibilities in blur-matching experiments. In a 2AFC staircase procedure, observers adjusted the blur of Gaussian edges (30% contrast) to match the perceived blur of various non-Gaussian test edges. In experiment 1, test stimuli were mixtures of 2 Gaussian edges (eg 10 and 30 min of arc blur) at the same location, while in experiment 2, test stimuli were formed from a blurred edge sharpened to different extents by a compressive transformation. Predictions of the various models were tested against the blur-matching data, but only one model was strongly supported. This was the template model, in which the input signature is the 2nd derivative of the luminance profile, and the templates are applied to this signature at the zero-crossings. The templates are Gaussian derivative receptive fields that covary in width and length to form a self-similar set (ie same shape, different sizes). This naturally predicts that shorter edges should look sharper. As edge length gets shorter, responses of longer templates drop more than shorter ones, and so the response distribution shifts towards shorter (smaller) templates, signalling a sharper edge. The data confirmed this, including the scale-invariance implied by self-similarity, and a good fit was obtained from templates with a length-to-width ratio of about 1. The simultaneous analysis of edge blur and edge location may offer a new solution to the multiscale problem in edge detection.

Contrast masking in strabismic amblyopia: attenuation, noise, interocular suppression and binocular summation

To investigate amblyopic contrast vision at threshold and above we performed pedestal-masking (contrastdiscrimination) experiments with a group of eight strabismic amblyopes using horizontal sinusoidal gratings (mainly 3 c/deg) in monocular, binocular and dichoptic configurations balanced across eye (i.e. five conditions). With some exceptions in some observers, the four main results were as follows. (1) For the monocular and dichoptic conditions, sensitivity was less in the amblyopic eye than in the good eye at all mask contrasts. (2) Binocular and monocular dipper functions superimposed in the good eye. (3) Monocular masking functions had a normal dipper shape in the good eye, but facilitation was diminished in the amblyopic eye. (4) A less consistent result was normal facilitation in dichoptic masking when testing the good eye, but a loss of this when testing the amblyopic eye. This pattern of amblyopic results was replicated in a normal observer by placing a neutral density filter in front of one eye. The two-stage model of binocular contrast gain control [Meese, T.S., Georgeson, M.A. & Baker, D.H. (2006). Binocular contrast vision at and above threshold. Journal of Vision 6, 1224--1243.] was `lesioned' in several ways to assess the form of the amblyopic deficit. The most successful model involves attenuation of signal and an increase in noise in the amblyopic eye, and intact stages of interocular suppression and binocular summation. This implies a behavioural influence from monocular noise in the amblyopic visual system as well as in normal observers with an ND filter over one eye.

Mach Bands: multi-scale spatial filtering and co-operative coding of edges and bars

Perception of Mach bands may be explained by spatial filtering ('lateral inhibition') that can be approximated by 2nd derivative computation, and several alternative models have been proposed. To distinguish between them, we used a novel set of ‘generalised Gaussian’ images, in which the sharp ramp-plateau junction of the Mach ramp was replaced by smoother transitions. The images ranged from a slightly blurred Mach ramp to a Gaussian edge and beyond, and also included a sine-wave edge. The probability of seeing Mach Bands increased with the (relative) sharpness of the junction, but was largely independent of absolute spatial scale. These data did not fit the predictions of MIRAGE, nor 2nd derivative computation at a single fine scale. In experiment 2, observers used a cursor to mark features on the same set of images. Data on perceived position of Mach bands did not support the local energy model. Perceived width of Mach bands was poorly explained by a single-scale edge detection model, despite its previous success with Mach edges (Wallis & Georgeson, 2009, Vision Research, 49, 1886-1893). A more successful model used separate (odd and even) scale-space filtering for edges and bars, local peak detection to find candidate features, and the MAX operator to compare odd- and even-filter response maps (Georgeson, VSS 2006, Journal of Vision 6(6), 191a). Mach bands are seen when there is a local peak in the even-filter (bar) response map, AND that peak value exceeds corresponding responses in the odd-filter (edge) maps.

The influence of fixation points on contrast detection and discrimination of patches of grating: Masking and facilitation

The use of fixation points (FPs) in visual psychophysics is common practice, though the costs and benefits of different fixation regimens have not been compared. Here we investigate the influence of several different types of FP configurations on the contrast detection of patches of sine-wave gratings. We find that for small targets (1°), the addition of a superimposed central FP can increase thresholds by a factor of 1.3 (2.5 dB) in comparison with no FP, and a factor of 1.5 (3.6 dB) in comparison with FPs that surround the target. These results are consistent with (i) a suppressive influence on the central region of the target from a central FP, and (ii) facilitatory influences from surrounding FPs. Our analysis of the slope of the psychometric function suggests that the facilitatory influence is not due to reduction of uncertainty. Plausible candidate causes for the facilitation are: (i) sensory interactions, (ii) aids to ocular accommodation and convergence, (iii) a reduction in eye-movements and (iv) more accurate placement of the observer’s window of attention. Masking by a central FP is not found for the suprathreshold task of contrast discrimination, suggesting that the masking effects of pedestal and FP do not combine linearly. This means that estimates of the level of masking produced by a contrast pedestal can depend on the details of the fixation point.

Size matters, but not for everyone:Individual differences for contrast discrimination

It is very well known that contrast detection thresholds improve with the size of a grating-type stimulus, but it is thought that the benefit of size is abolished for contrast discriminations well above threshold (e.g., Legge, G. E., & Foley, J. M. (1980)]. Here we challenge the generality of this view. We performed contrast detection and contrast discrimination for circular patches of sine wave grating as a function of stimulus size. We confirm that sensitivity improves with approximately the fourth-root of stimulus area at detection threshold (a log-log slope of -0.25) but find individual differences (IDs) for the suprathreshold discrimination task. For several observers, performance was largely unaffected by area, but for others performance first improved (by as much as a log-log slope of -0.5) and then reached a plateau. We replicated these different results several times on the same observers. All of these results were described in the context of a recent gain control model of area summation [Meese, T. S. (2004)], extended to accommodate the multiple stimulus sizes used here. In this model, (i) excitation increased with the fourth-root of stimulus area for all observers, and (ii) IDs in the discrimination data were described by IDs in the relation between suppression and area. This means that empirical summation in the contrast discrimination task can be attributed to growth in suppression with stimulus size that does not keep pace with the growth in excitation. © 2005 ARVO.

Neuromagnetic evoked responses to complex motions are greatest for expansion

We analysed evoked magnetic responses to moving random dot stimuli, initially using a 19-channel magnetoencephalography (MEG) system, and subsequently using a 151-channel MEG system. Random dot displays were used to construct complex motion sequences, which we refer to as expansion, contraction, deformation, and rotation. We also investigated lateral translation and a condition in which the directions of the dots were randomised. In all stimulus conditions, the dots were first stationary, then traveled for a brief period (317s or 542 ms), and were then stationary again. In all conditions, evoked magnetic responses were observed with a widespread bilateral distribution over the observers' heads. Initial recordings revealed a substantially larger evoked magnetic response to the expansion condition than the other conditions. In a revised study, we used a 151-channel MEG system and two stimulus diameters (9.3 and 48 deg), the smaller comparable with the first experiment. The responses were analysed using a nonparametric approach and confirmed our initial observations. In a third study, speed gradients were removed and a new design permitted direct comparisons between motion conditions. The results from all three experiments are consistent with the greater ecological validity of the expansion stimulus. © 2004 Elsevier B.V. All rights reserved.

Area summation and masking

At detection threshold, sensitivity improves as the area of a test grating increases, but not when the test is placed on a pedestal and the task becomes contrast discrimination (G. E. Legge, & J. M. Foley, 1980). This study asks whether the abolition of area summation is specific to the situation where mask and test stimuli have the same spatial frequency and orientation ("within-channel" masking) or is more general, also occurring when mask and test stimuli are very different ("cross-channel" masking). Threshold versus contrast masking functions were measured where the test and mask were either both small (SS), both large (LL), or small and large, respectively (SL). For within-channel masking, facilitation and area summation were found at low mask contrasts, but the results for SS and LL converged at intermediate contrasts and above, replicating Legge and Foley (1980). For all three observers, less facilitation was found for SL than for SS. For cross-channel masking, area summation occurred across the entire masking function and results for SS and SL were identical. The results for the entire data set were well fit by an extended version of a contrast masking model (J. M. Foley, 1994) in which the weights of excitatory and suppressive surround terms were free parameters. I conclude that (i) there is no empirical abolition of area summation for cross-channel masking, (ii) within-channel area summation can be abolished empirically without being disabled in the model, (iii) observers are able to restrict the area of spatial integration, but not suppression, (iv) extending a cross-channel mask to the surround has no effect on contrast detection, and (v) there is a formal similarity between area summation and contrast adaptation. © 2004 ARVO.

Extrinsic factors in the perception of bistable motion stimuli

When viewing a drifting plaid stimulus, perceived motion alternates over time between coherent pattern motion and a transparent impression of the two component gratings. It is known that changing the intrinsic attributes of such patterns (e.g. speed, orientation and spatial frequency of components) can influence percept predominance. Here, we investigate the contribution of extrinsic factors to perception; specifically contextual motion and eye movements. In the first experiment, the percept most similar to the speed and direction of surround motion increased in dominance, implying a tuned integration process. This shift primarily involved an increase in dominance durations of the consistent percept. The second experiment measured eye movements under similar conditions. Saccades were not associated with perceptual transitions, though blink rate increased around the time of a switch. This indicates that saccades do not cause switches, yet saccades in a congruent direction might help to prolong a percept because i) more saccades were directionally congruent with the currently reported percept than expected by chance, and ii) when observers were asked to make deliberate eye movements along one motion axis, this increased percept reports in that direction. Overall, we find evidence that perception of bistable motion can be modulated by information from spatially adjacent regions, and changes to the retinal image caused by blinks and saccades.

Response normalization and blur adaptation:data and multi-scale model

Adapting to blurred or sharpened images alters perceived blur of a focused image (M. A. Webster, M. A. Georgeson, & S. M. Webster, 2002). We asked whether blur adaptation results in (a) renormalization of perceived focus or (b) a repulsion aftereffect. Images were checkerboards or 2-D Gaussian noise, whose amplitude spectra had (log-log) slopes from -2 (strongly blurred) to 0 (strongly sharpened). Observers adjusted the spectral slope of a comparison image to match different test slopes after adaptation to blurred or sharpened images. Results did not show repulsion effects but were consistent with some renormalization. Test blur levels at and near a blurred or sharpened adaptation level were matched by more focused slopes (closer to 1/f) but with little or no change in appearance after adaptation to focused (1/f) images. A model of contrast adaptation and blur coding by multiple-scale spatial filters predicts these blur aftereffects and those of Webster et al. (2002). A key proposal is that observers are pre-adapted to natural spectra, and blurred or sharpened spectra induce changes in the state of adaptation. The model illustrates how norms might be encoded and recalibrated in the visual system even when they are represented only implicitly by the distribution of responses across multiple channels.

The effect of inversion on the encoding of normal and “thatcherized” faces

In the "Thatcher illusion" a face, in which the eyes and mouth are inverted relative to the rest of the face, looks grotesque when shown upright but not when inverted. In four experiments we investigated the contribution of local and global processing to this illusion in normal observers. We examined inversion effects (i.e., better performance for upright than for inverted faces) in a task requiring discrimination of whether faces were or were not "thatcherized". Observers made same/different judgements to isolated face parts (Experiments 1-2) and to whole faces (Experiments 3-4). Face pairs had the same or different identity, allowing for different processing strategies using feature-based or configural information, respectively. In Experiment 1, feature-based matching of same-person face parts yielded only a small inversion effect for normal face parts. However, when feature-based matching was prevented by using the face parts of different people on all trials (Experiment 2) an inversion effect occurred for normal but not for thatcherized parts. In Experiments 3 and 4, inversion effects occurred with normal but not with thatcherized whole faces, on both same- and different-person matching tasks. This suggests that a common configural strategy was used with whole (normal) faces. Face context facilitated attention to misoriented parts in same-person but not in different-person matching. The results indicate that (1) face inversion disrupts local configural processing, but not the processing of image features, and (2) thatcherization disrupts local configural processing in upright faces.