Our brains continuously transform sensory information into meaningful perceptions. We try to understand the brain processes involved in visual transformations, such as encoding sequences of line points into curves, curves into shapes, and shapes into recognizable images. Though these processes are mostly visual, we find they also make use of more general brain functions, such as information chunking, learning and memory, that are employed by other brain faculties. Transforming light into images may not be a very different task from transforming sound waves into music or ideas into thoughts.

Toward achieving the goal of understanding human vision we use psychophysical methods, in an attempt to quantify perceptual and cognitive abilities. Though human brain is not yet readily accessible to direct activity measurements, much of its logic can be uncovered by measuring human performance in well controlled settings. For example, our inability to discriminate between some color mixtures (red+green = yellow) puts constraints on models of color processing, and detailed experiments can be carried out to further understand the way we see colors. In a similar way, we try to understand processes involved in pattern vision, by manipulating specific shape components. We design computer generated displays aimed at testing human performance on well defined detection and discrimination tasks, with visual targets carefully designated to probe brain processes such as image segmentation, perceptual organization, learning and mental imagery.

As our visual system is constructed from many interacting modules, each dealing with a different aspect of the visual task, we made a strategic decision to start from relatively simple low level processes involved in image segmentation. These processes, probably residing at the entrance stage of the visual cortex, were believed to be devoid of cognitive intervention, and indeed we could successfullymodel performance on texture segmentation and perceptual grouping tasks by using simple localized image-analyzers with lateral excitatory and inhibitory interactions. The architecture of these interactions was explored using contrast detection tasks. However, it became evident that performance on these texture tasks improves with time, pointing toward learning effects. Further experiments provided evidence of a genuine learning process, probably governed by associative rules, occuring at an early stage of visual processing. Our extended knowledge of segmentation processes contributed here to an understanding of the learning process and to quantify some learning abilities. Recently we have also demonstrated cognitive modulation of lateral interactions by usingmental imagery (as in trying to imagine a visual object).

The research described above is being carried out in collaboration with young scientists, working toward their doctoral degree. The Weizmann Institute and the Center for the study of Higher Brain Function produce an excellent environment for them to express their original ideas and to mature as excellent scientists. Other collaborative efforts are made with Neurologists at the Sheba Medical Center (Tel-Hashomer) and at the Loewenstein Rehabilitation Hospital (Raanana) to better understand brain damage affecting vision. I received my Ph.D. in Neurobiology from the Hebrew University (Jerusalem) and further training at the AT&T Bell Laboratories and at the California Institute of Technology.

Rubenstein, B. S. and Sagi, D. (1990): Spatial variability as a limiting factor in texture discrimination tasks: Implication for performance asymmetries. Journal of the Optical Society of America A, 7, 1632-1643.

Karni, A. and Sagi, D. (1993): The time course of learning a visual skill. Nature, 365, 250-252.

Polat, U. and Sagi, D. (1994): Spatial interactions in human vision: from near to far via experience dependent cascades of connections.Proceed. of the Nat. Acad. of Sci. USA, 91, 1206-1209.

Barchilon Ben-Av, M. and Sagi, D. (1995): Perceptual grouping by similarity and proximity: Experimental results can be predicted by intensity autocorrelations. Vision Research, 35, 853-866.

Ishai A. and Sagi, D. (1995) Common Mechanisms of Visual Imagery and Perception. Science, 268, 1772-1774.

Click here for the Psychophysics Lab members .  

Tel: +972(8)934-3747

Fax: +972(8)934-4131

e-mail: Dov.Sagi@Weizmann.ac.il

 

 

Images used to explore perceptual organization. The display elements are arranged to produce vertical grouping based on elements similarity in A, horizontal grouping based on elements proximity in B, and ambiguous organization when similarity cues are put in conflict with proxmity cues in C. Human performance on such grouping tasks can be predicted by intensity autocorrelations.

---

This site was developed by Chaipi Wijnbergen using Accent Multilingual Publisher on 2/4/96

For more information send mail to Ariela Sharfi