Learning to Segment Images Using Dynamic Feature Binding

Mozer, Michael C.; Zemel, Richard S.; Behrmann, Marlene

doi:10.1162/neco.1992.4.5.650

Cited by 62 publications

(48 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We hypothesize that a grouping process is an essential component of object-based attention because these grouped features are gated by a selection attention mechanism. In our previous work with MAGIC ( Mozer et al, 1992 ), our concern was more with the nature of the phase alignment. Our concern, here, however, is to provide a framework within which to interpret the empirical findings.…”

Section: A Decision Processmentioning

confidence: 99%

Object-based attention and occlusion: Evidence from normal participants and a computational model.

Behrmann¹,

Zemel²,

Mozer³

1998

Journal of Experimental Psychology: Human Perception and Perfor

Self Cite

175

203

View full text Add to dashboard Cite

One way of perceptually organizing a complex visual scene is to attend selectively to information in a particular physical location. Another way of reducing the complexity in the input is to attend selectively to an individual object in the scene and to process its elements preferentially. This latter, object-based attention process was examined, and the predicted superiority for reporting features from 1 relative to 2 objects was replicated in a series of experiments. This object-based process was robust even under conditions of occlusion, although there were some boundary conditions on its operation. Finally, an account of the data is provided via simulations of the findings in a computational model. The claim is that object-based attention arises from a mechanism that groups together those features based on internal representations developed over perceptual experience and then preferentially gates these features for later, selective processing. Humans are exceptionally good at recognizing objects in natural visual scenes despite the fact that such scenes usually contain multiple, overlapping objects. One way in which individuals organize this complex input to minimize the processing load is to divide the field on the basis of spatial location and then to attend selectively to particular physical regions. This selective attentional spotlight "illuminates" areas of interest and facilitates preferential processing of information from those chosen areas (e.g., Broadbent, 1982 ;B. A. Eriksen & Eriksen, 1974 ;C. W. Eriksen & Yeh, 1985 ;Posner, 1980 ). There is now much evidence supporting this location-based selection, all of which shows that information from selected regions is processed faster and more accurately than equivalent information from unattended regions ( Posner, 1980 ;Posner, Snyder, & Davidson, 1980 ). The idea that location-based selection plays an exclusive role in organizing visual information, however, has been increasingly challenged in recent years. Studies have shown, for example, that humans can select one of two superimposed figures even when there is no spatial basis for selection ( Rock & Gutman, 1981 ) and can allocate attention to perceptual groups independent of the spatial proximity and contiguity of the component elements (e.g., Behrmann, Vecera, & McGoldrick, 1998 ;Driver & Baylis, 1989 ;Duncan, 1984 ;Kramer & Jacobson, 1991 ;Kramer & Watson, 1995 ;Lavie & Driver, 1996 ;Prinzmetal, 1981 ;Vecera & Farah, 1994 ). To account for these findings, an alternative selection process, in which attention is directed to objects, rather than to locations or unsegmented regions of space, has been proposed. This object-based mechanism, in which complex visual input is parsed into discrete units for further processing, has received considerable empirical, neuropsychological, and computational support in recent years. Object-Based Visual AttentionAn early but compelling empirical illustration of the view that attention can be directed to objects, rather than to spatial locations per se, comes...

show abstract

Section: A Decision Processmentioning

confidence: 99%

Object-based attention and occlusion: Evidence from normal participants and a computational model.

Behrmann¹,

Zemel²,

Mozer³

1998

Journal of Experimental Psychology: Human Perception and Perfor

Self Cite

175

203

View full text Add to dashboard Cite

show abstract

“…However, knowledge could also be stored at the same level of representation as the image. Local conjunctions of features could, in principle, be learned and stored in the connections among the feature representations in the image.This is essentiallyhow Mozer, Zemel, Behrmann, & Williams, 1992) model of image segmentation codes statistical information contained in images.…”

Section: Knowledge In the Visual Systemmentioning

confidence: 99%

“…Mozer and his colleagues Mozer, Zemel, Behrmann, & Williams, 1992) trained a connectionist network to segment images consisting of two overlapping objects, say, two squares. The significant contribution of this work is that it exploits the ability of connectionist networks to learn, and the network discovers grouping principles instead of having to have the heuristics built in by the programmer.…”

mentioning

confidence: 99%

Is visual image segmentation a bottom-up or an interactive process?

Vecera

Farah

1997

Perception & Psychophysics

144

102

View full text Add to dashboard Cite

Visualimage segmentation is the process by which the visual system groups features that are part of a single shape. Is image segmentation a bottom-up or an interactive process? In Experiments 1 and 2, we presented subjects with two overlapping shapes and asked them to determine whether two probed locations were on the same shape or on different shapes. The availability of top-down support was manipulated by presenting either upright or rotated letters. Subjects were fastest to respond when the shapes corresponded to familiar shapes-the upright letters. InExperiment 3, we used a variant of this segmentation task to rule out the possibility that subjects performed same/different judgments after segmentation and recognition of both letters. Finally,in Experiment 4, we ruled out the possibility that the advantage for upright letters was merely due to faster recognition of upright letters relative to rotated letters. The results suggested that the previous effects were not due to faster recognition of upright letters; stimulus familiarity influenced segmentation per se. The results are discussed in terms of an interactive model of visual image segmentation.Imagine a rider atop a horse. There are clearly points ofoverlap between the two objects, such as where the rider's legs occlude the side ofthe horse. For the visual system to recognize these two distinct objects, each visual object representation stored in visual memory must receive input from the regions of the visual field that correspond to that object and only to that object. Visual image segmentation is the process by which the visual system groups or "binds" locations and features that are part of a single shape. Is image segmentation a purely bottomup, stimulus-driven process or can stored knowledge of objects playa role in image segmentation?Both answers seem equally plausible. Bottom-up image segmentation is consistent with Fodor's (1983) claims of information encapsulation in the peripheral or input sys-

show abstract

“…Although DUBM units contain a rotational variable, this is not used to model relative rotations of subcomponents. For example in [15] the authors present a convolutional architecture where the rotational variable denotes the phase of an oscillator, relating to the theory of binding-by-synchrony.…”

Section: Related Workmentioning

confidence: 99%

Transformation Equivariant Boltzmann Machines

Kivinen

Williams

2011

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Abstract. We develop a novel modeling framework for Boltzmann machines, augmenting each hidden unit with a latent transformation assignment variable which describes the selection of the transformed view of the canonical connection weights associated with the unit. This enables the inferences of the model to transform in response to transformed input data in a stable and predictable way, and avoids learning multiple features differing only with respect to the set of transformations. Extending prior work on translation equivariant (convolutional) models, we develop translation and rotation equivariant restricted Boltzmann machines (RBMs) and deep belief nets (DBNs), and demonstrate their effectiveness in learning frequently occurring statistical structure from artificial and natural images.

show abstract

Learning to Segment Images Using Dynamic Feature Binding

Cited by 62 publications

References 20 publications

Object-based attention and occlusion: Evidence from normal participants and a computational model.

Object-based attention and occlusion: Evidence from normal participants and a computational model.

Is visual image segmentation a bottom-up or an interactive process?

Transformation Equivariant Boltzmann Machines

Contact Info

Product

Resources

About