Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Gestalt Issues in Modern Neuroscience Walter H. Ehrenstein, Lothar Spillmann and Viktor Sarris Axiomathes 13: 433–458, 2003. © 2003 Kluwer Academic Publishers. Printed in the Netherlands. Thomas E. Shepherd [email protected] \\Dbmn6x21\Documents\cse7521 Quick Overview of Gestalt Theory Gestalt theory is a family of psychological theories, that have influenced many research areas since 1924. contents of paper 1. PHENOMENOLOGY AS A GUIDE TO BRAIN RESEARCH 2. PURPOSE AND PROCESSING MODES OF PERCEPTION 3. RECEPTIVE FIELD AS A FUNCTIONAL MICROGESTALT 4. RECEPTIVE AND PERCEPTIVE FIELD ORGANIZATION 5. GENERAL LINKING PROPOSITIONS: ISOMORPHISM 6. FURTHER ISSUES OF GESTALT PSYCHOPHYSICS 7. EPILOGUE 1. PHENOMENOLOGY AS A GUIDE TO BRAIN RESEARCH This paper presents a set of examples of how visual phenomena can serve as tools to uncover brain mechanisms. Specifically, receptive field organization is proposed as a Gestalt-like neural mechanism of perceptual organization. 2. PURPOSE AND PROCESSING MODES OF PERCEPTION A key assumption of Gestalt psychology is that percepts organize themselves in the simplest, most regular and balanced manner possible under the prevailing stimulus conditions. What is the purpose of this tendency towards organization? A possible answer is optimization in the interest of robust transmission of information. Straight lines, continuous contours, symmetrical shapes are ubiquitous properties of natural objects. 3. RECEPTIVE FIELD AS A FUNCTIONAL MICRO-GESTALT The neuronal mechanism that encodes optical contrast is the receptive field organization. This is the area of the retina within which a change of luminance induces a change of the neuronal response. Receptive fields are • subdivided into a circular center and a concentric surround. • the center and the surround are antagonistically organized, • encode spectral or chromatic stimuli • are structured for selective spatio-temporal sampling The hypothesis : receptive field organization is a basic mechanism that allows for non-additive, Gestalt-like integration of visual input. 3.1. Receptive and perceptive fields A and well-known phenomenon that may be explained in terms of concentric receptive fields is the Hermann grid illusion. 4. RECEPTIVE AND PERCEPTIVE FIELD ORGANIZATION A major change in receptive field structure occurs at the cortical level. retinal receptive fields have a circular center-surround structure cortical receptive fields are elongated, as if formed by collinearly overlapping receptive fields receptive fields increase in size as they are abstracted from the retina Context neurons The assumption is that Gestalt factors inherent in perceptual organization may be based on special context neurons, i.e., on cells that are sensitive to the context of the stimulus presented. Neurons have been found that respond differentially to local and global stimuli. For example : a hatched square will elicit different neuronal responses depending on whether it is placed on a cross or iso-hatched background. There are neurons in monkey area V2 responding to lateral disparity, i.e., to the small differences between the two monocular images : Stereo perception ... Edge-polarity neurons Figures on a ground are privileged by being surrounded by a ‘unilateral’ border that separates them from the ground. The figure “owns” the border or else the border “belongs” to the figure, not to the ground. edge polarity neurons in monkey areas V2 and V4 that respond to a lightdark step in one direction, but not in the other. Given the right direction of polarity this asymmetric response in conjunction with a closed contour could underlie “belongingness”. End-stopped units and neurons signaling subjective contours The same kind of mechanism may explain the illusory contours in the previous Kanizsa triangle and the Ehrenstein circle. Baumgartner et al. (1984) showed that the response of a neuron in area V2 to an interrupted bar was qualitatively similar, although weaker, than the response of the same neuron to a continuous bar. Signal propagation from the edge : Filling-in Much of the stimulus information generated by a visual figure is contained in the contour, but we also have access to uniform brightness, color, and texture that fill the enclosed surface area. From observations using strict visual fixation we know that surfaces tend to become leveled relative to the ground and will even fade from view, due to local adaptation. The main mechanism to overcome adaptation, and to sustain the perception of a surface over time is the continuous update of signals emanating from the edges. On average, involuntary saccades will revive a percept four times a second thereby preventing its disappearance in the background Component cells and pattern neurons : Local and global analysis of motion The perceived direction of moving contours depends on their spatial context, i.e., on the given shape of an “aperture” within which the moving This phenomenon refers to a general problem of motion analysis : How are the individual directions of local motion signals integrated so that they result in global motion events within a complex visual scene? Neurophysiologically, the visual system faces the same problem in that limitations arise from the given shape and size of the receptive field of a single neuron Synchronized neural activity: Common fate A mechanism that may achieve perceptual integration over extended areas of the visual field may be the temporal binding of neuronal activity. Consider a figure composed of a number of dots on a randomly dotted background. As long as the figure remains static, it is embedded within the background dots and is thus invisible. However, as soon as it moves the figure pops out testifying to the enormous power of coherent motion. The Gestaltists called this principle of grouping the factor of common fate; it can tie together objects that are quite distant and different in form, size, or color. GENERAL LINKING PROPOSITIONS : ISOMORPHISM Are there general rules of how visual phenomena are related to brain function? Gestalt isomorphism postulates that every perceptual state is linked to a structurally identical neural process in the observer’s the brain. The concept of Gestalt isomorphism has not yet (?) received much attention in neuroscience. FURTHER ISSUES OF GESTALT PSYCHOPHYSICS The brain processes underlying many of the phenomena related to sensory and memory functions are still to be identified. They likely involve integrated neuronal networks rather than local cell properties. Frame of reference and relational psychophysics The concept of the frame of reference refers to the fact that identical stimuli can appear differently depending on the surround conditions. The ratio principle refers to the fact that a perceived quality is not locally determined, but is related to every other part of the visual field. For instance, the perception of surface lightness is based on the ratio between the light from the figure to that from its immediate surround Geometrical optical distortions The Ebbinghaus illusion belongs to the “mixed” type of geometric-optical distortions in that both the distance (D) and the surrounding context-size (B) may be systematically varied together with the focus stimulus (X) to be judged. A digression into Chickens ... The phenomenon of transposition refers to the fact that perception preserves the relations between stimuli, despite large changes in their absolute level. As an example, after a chicken is trained to respond to the larger of a pair of images, when presented with a new pair, it will typically select the larger. A two layer neural network was developed and trained to simulated a range of perceptual learning phenomena chickens. 7. EPILOGUE We have argued in favor of a wider scope of phenomenology, traditionally resident in philosophy and psychology, to enrich the perspectives of modern “Gestalt psychophysics” and brain research. Perhaps we should target image generation, not at the level of retinal stimulation, but toward a “higher” level of Gestalt stimulation ... ?