Visual masking: past accomplishments, present status, future developments

Visual masking, throughout its history, has been used as an investigative tool in exploring the temporal dynamics of visual perception, beginning with retinal processes and ending in cortical processes concerned with the conscious registration of stimuli. However, visual masking also has been a phenomenon deemed worthy of study in its own right. Most of the recent uses of visual masking have focused on the study of central processes, particularly those involved in feature, object and scene representations, in attentional control mechanisms, and in phenomenal awareness. In recent years our understanding of the phenomenon and cortical mechanisms of visual masking also has benefited from several brain imaging techniques and from a number of sophisticated and neurophysiologically plausible neural network models. Key issues and problems are discussed with the aim of guiding future empirical and theoretical research.

important in the studying the microgenesis of object perception. I cannot review all of the related accomplishments of the past. For that I refer the reader to Chapter 1 of the 2 nd edition of our book, Visual Masking . It amply reviews the history of masking from the late 19 th century to the middle of the 20 th . Looking at the wider span of about 140 years up to the present, one can, however, discern some interesting features, transitions, or phases in the study of masking. Toward the turn of the 19 th century, masking was viewed as a way of exploring interactions thought to occur anywhere along the visual tract, from lateral interactions in the retina to cortical processes underlying object cognition and consciousness. With the ascendance of behaviorism some decades later, the topic of cognition and especially consciousness took a nosedive toward oblivion.
With the exception of Piéron's (1935) and Werner's (1935) Crawford (1947) and on metacontrast by Alpern (1953) toward the middle of the 20 th century. Both investigations and their immediate offshoots focused on pro-cesses -early light and dark adaptation, interactions among rod and cone activations -that were deemed to occur at early, peripheral levels. Neither was remotely concerned with higher brain processes related to cognition or consciousness. While masking by light is largely confined to peripheral, most likely retinal, processes (Battersby, Oesterreich, & Sturr, 1964), we now know that the crucial aspects of metacontrast and pattern masking are determined by cortical interactions. Since the http://www.ac-psych.org Bruno G. Breitmeyer 1960s very few studies were conducted on masking by light, and none that I know of since Cogan's (1989Cogan's ( , 1992) studies in the late 1980s and early 1990s. In contrast, pattern masking and metacontrast studies retained their currency to the present. Why?
I believe three trends in scientific outlook merged mid century to promote continued interest in, among many other topics, pattern masking. Because they specify and actualize a single or a few constellations of features from among a vastly larger set of possibilities, patterns are organized physical or mental entities that convey information. Within that context, one trend was the theory of communication (Shannon & Weaver, 1948), which formalized a rigorous mathematical definition of information in terms of bits. In turn this formalization could be wedded readily with a second concurrent formalization in computational science and artificial intelligence (Turing, 1950). The third was the pioneering work of Hebb (1949) attempting to reconcile phenomenological Gestalt and functional "connectionist" approaches in a plausible neural-network model of the organization of mind and its perceptual and cognitive control of behavior. The imprint of the former influence was clearly left on the pioneering works of Cherry (1953), Broadbent (1958) and Moray (1959) on the role and properties of attention in various "capacity-limited" sensory "channels" of communication, and with respect to masking on the information-processing approaches to visual cognition, with all its "parallel" and "serial" processors, adopted from the early 1960s through 1970s by Averbach and Coriell (1961), Sperling (1963), Scheerer (1973), and Turvey (1973. Additionally, in the late 1950s and early 1960s artificial intelligence spurred, among other things, development of computational models of perception and pattern recognition such as Rosenblatt's (1958) Perceptron and Selfridge's (1959;Selfridge & Neisser, 1960) Pandemonium. And Hebb's (1949) related work on physiologically plausible neural networks of perception anticipated the first attempts around 1970 at providing quantitative neural network models of pattern masking by Weisstein (1968) and by Bridgeman (1971). What I consider to be an important transitional approach to masking was the work of Bachmann (1984Bachmann ( , 1994, which appeared at about the same time as the first edition of my book on visual masking highlighting the dual-channel, sustained-transient approach to masking (Breitmeyer & Ganz, 1976

direct parameter specification and masked priming
In the late 1980s and early 1990s, a new methodological application of metacontrast masking evolved in the context of the theory of direct parameter specification (DPS). Formulated by the Bielefeld group under the direction of Odmar Neumann, DPS took the findings originally reported by Fehrer and Raab (1962), that a fully masked target could activate processes that facilitated response times in a simple detection task, one step further by arguing and showing that a suppressed target could additionally prime sensori-motor pathways specified by  (Breitmeyer, Öğmen, & Chen, 2004;Breitmeyer, Ro, & Singhal, 2004;Breitmeyer, Öğmen, Ramon, & Chen, 2005). More on that later.

four-dot and common-onset masking
During the 1993 meeting of the Psychonomics Society held in Washington, D. C., I had the pleasure of exchanging ideas with Vince Di Lollo on several occasions. On one occasion Vince enthusiastically described the four-dot and common-onset masking techniques (Bischof & Di Lollo, 1995;Di Lollo, Bischof, & Dixon, 1993) and their implications for -in his terms -a fundamentally new conceptualization of masking in terms of downward influences from higher-level processes instead of low-level contour interactions.
I was skeptical and privately dismissed his enthusiasm as heady overexcitement. After all, I thought, Naomi Weisstein, Charlie Harris, and their collaborators (Weisstein & Harris,1974;Williams & Weisstein, 1978 (Edelman, 1987;Posner, 1994;Zeki, 1993) is also a central theme in the theory of objectsubstitution masking (Enns, 2004;Enns & Di Lollo, 1997;Di Lollo, Enns, & Rensink, 2000); and I will argue later that it also will have to be incorporated into other neural network models that make claims to physiological realism. Just as Bachmann's model of perceptual retouch (PR) -which by the way is a form of object substitution -placed the spotlight on the underadvertised existence of the retino-reticular-thalamic activations, so does object-substitution masking highlight the important roles of heretofore underadvertised yet massive reentrant pathways in the cortical visual system. More on that later also.

Neuroscientific approaches to masking
The first neuro-and electrophysiological studies of masking go back nearly four decades. I will not review all of the studies that have been conducted since then; such a review is found in Chapter 3 of our forthcoming book on visual masking . I will highlight the few that, in my opinion, are most revealing in relation to metacontrast and para-contrast masking. Of the older studies, the studies by Schiller and Chorover (1966), Vaughn and Silverstein (1968), and Schwartz and Pritchard (1981) recording human cortical visual evoked potentials (CVEPs) and Bridgeman's (1980) studies of single cortical cells in monkey all indicate that it is the variations of the later response components of the V1 cortical response which correlate with visibility of a target during metacontrast. When I read these studies, I took their results as confirming the sustained-transient channel approach to masking. According to that model, one would expect suppression of cortical responses to occur in the longer-latency sustained channels, which I assumed were responsible for generating the longer latency or late CVEP components. In gist I believe this is still correct, but not in detail. The reason is that the original dual-channel approach was developed within a feedforward framework. More recent neurophysiological results, however, seriously question this framework.
According to Lamme and coworkers (Lamme, 1995;Lamme, Super, Landman, Roelfsema, & Spekreijse, 2000;Super, Spekreijse, & Lamme, 2001), the late V1 response component, as shown in Figure 1, is associated with figure 1. Lamme, Super, Landman, Roelfsema, & Spekreijse, 2000)  I believe this view is also consistent with the some of the recent results reported by Macknik and Livingstone (1998). They showed (see Figure 2) that metacontrast suppresses a later target-response component which they associated with the offset of the target, whereas it had virtually no effect on the early response compo- Based on their results and on the above reasoning, Macknik and Livingstone (1998) developed what I believe to be currently the most effective masking method, namely, the standing-wave illusion, for rendering stimuli invisible. In this method a mask appears about 100 ms before the target, which in turn is followed about 50 ms by the mask, followed 100 ms by the target and so on. Basically the target and mask are presented at optimal para-and metacontrast SOAs throughout the presentation (see Figure 5 below), thus giving the target a "double masking whammy" by suppressing first its feedforward activity and then in addition the (already weakened) re-entrant activity.

neural-network modeling
For these reasons I maintain that neural-network models of backward pattern masking need to pay due attention to re-entrant cortical activations. Our updated REtinalCOrticalDynamics (RECOD) model , which Haluk Öğmen will cover more extensively, incorporates (a) Comparison of a typical masking function obtained in our laboratory using a visual para-or metacontrast mask and a typical masking function obtained by Corthout, Uttl, Ziemann et al. (1999) using a TMS pulse as a mask. Negative and positive SOAs indicate that the masks were presented before and after the target, respectively. Results are not adjusted for retinocortical transmission delay. (b) Same as preceding but with results adjusted for a 60-ms delay of cortical M activity due to retinocortical transmission time (Baseler & Sutter, 1997). (From Breitmeyer, Ro, Öğmen, 2004) Unadjus http://www.ac-psych.org building. One finding is the very existence of commononset masking (Bischof, & Di Lollo, 1995;Di Lollo, Bischof, & Dixon, 1993;Di Lollo, Enns, & Rensink, 2000). Of course this finding is explained by the OS model. I think Bachmann's PR model might also give an adequate account of the major aspects of commononset-masking. While it has been suggested that some former models such as Bridgeman's Hartline-Ratliff neural net may also give an account of common-onset masking (Bischoff & Di Lollo, 1995), Greg Francis's recent work (Francis & Cho, 2006, submitted)  In another study (Breitmeyer, Kafaligönül, Öğmen, Mardon, Todd, & Ziegler, 2006), we also have shown that metacontrast masking can separately affect con-tour and surface properties of visual objects. In this study, observers were required to judge the target either with regard to its contour detail or else its surface brightness. The results, shown in Figure 7, show that two distinct metacontrast functions are obtained for these two correspondingly distinct tasks. Both tasks yielded typical U-shaped metacontrast functions. However, while the contour task yielded optimal masking at a short SOA of 10 ms, the brightness task yielded optimal masking at a higher SOA of 40 ms.
This indicates that an object's surface brightness is processed about 30 ms later than its contour. These findings are consistent with several theoretical and empirical results. For one, Grossberg and colleagues (Cohen & Grossberg 1984;Grossberg 1994;Grossberg & Yazdankbakhsh, 2005)  Metacontrast contour and surface-contrast suppression as a function of stimulus onset asynchrony (SOA). (Adapted after  http://www.ac-psych.org is assumed to be intimately tied to attention (Enns & Di Lollo, 1997;Di Lollo et al., 2000), this featurespecific OS masking is entirely consistent with other recent reports of feature-based (as compared to object-based) attention (Hayden & Gallant, 2005;Nobre, Rao & Chelazzi, 2006) In view of these findings, I think that a clear theoretical statement specifying the relation between features and objects may need to be spelled out in the OS model.

WHat next?
As with weather forecasting, forecasting developments in any field of research is an inexact exercise. The safest bet is that things will be much the same tomorrow as today. Easier is the task of posing questions that might define some of the paths that future developments take. antagonism of classical receptive fields not only at cortical levels but also at subcortical levels, as originally proposed by Breitmeyer and Ganz (1976). Since the surround response lags the center response by 10-30 ms, one would expect optimal paracontrast at a very short negative SOA. Figure 8 shows a typical result from a recent studies  conducted in our laboratories. Here a contour discrimination task was used to index masking. Note that indeed a local maximum in the masking effect occurs at an SOA of -10 ms. This would be consistent with center-surround interactions within antagonistically organized receptive fields. However, note also that there is a second maximal masking effect at an SOA of roughly 200 ms, more in line with neurophysiological findings reported by Macknik and Livingstone (1998) and with prior psychophysical findings (Cavonius & Reeves, 1983;Scharf & Lefton, 1970 masking. While these are all useful ways of "skinning" consciousness, they do not yield equivalent results. Figure 9 shows results we (Breitmeyer, Öğmen, & Koç, 2005) recently obtained in which metacontrast masking was studied under nonrivalrous dichoptic viewing in comparison to when the eye to which the mask was presented was in the suppressed phase of binocular rivalry. Note that in the nonrivalrous condition, the results indicate low visibility of the target and high visibility of the mask, a result typical under standard dichoptic viewing of the stimuli (Kolers & Rosner, 1960;Schiller & Smith 1968, Weisstein, 1971. However, in the rivalrous condition, the target's visibility is no longer suppressed, while that of the mask is. This target recovery or disinhibition in the rivalrous condition indicates that not only the neural processes responsible for the visibility of the mask but also those responsible for its effectiveness as a suppressor of the target are suppressed during binocular rivalry. In other words, here we do not obtain the aforementioned dissociation between the two distinct mask-activated neural processes. This indicates that binocular-rivalry can suppress the metacontrast mechanism and thus that binocular-rivalry suppression and metacontrast suppression work at different functional levels of processing. In some sense binocular-rivalry suppression is functionally prior to metacontrast suppression. How this might translate into underlying neurophysiology is hard to assess. However, at first glance the priority of binocular-rivalry relative to metacontrast suppression appears consistent with a) the results reported by Martinez-Conde (2004), Haynes Deichmann, and, and Tse et al. (2005) showing that metacontrast and visual pattern masking occur at fairly late levels in the cortical visual pathway and 2) the recent findings showing neural signatures of binocular rivalry suppression in humans as early as the lateral geniculate nucleus (Haynes, Deichmann et al., 2005, Wunderlich, Schneider, & Kastner, 2005. For these reasons, I believe that by looking at how masking relates to other psychophysical "blinding" methods and how any emerging differences correlate with differ-