The Detection and Segmentation of Image Motion

Ohio State University, The, Columbus OH

Investigators

Abstract

A fundamental problem for both human and machine is to detect and identify moving patterns from sensory information that is often contaminated by noise. Previous research has shown that one way of enhancing signal to noise sensitivity is to sum the outputs of many different motion detectors in different locations of the visual field. An important difficulty with this approach, however, is how to decide which local motion detectors should be summated. Whereas previous studies have examined the case of simple translatory motion where all elements in a pattern move at the same velocity, our research will consider more complicated motions in which it is possible for patterns to rotate or expand over time. We will also investigate the detection of camouflaged objects whose motions are confined to a limited region of the visual field. A series of psychophysical experiments will be performed to isolate the basic mechanisms by which the human visual system is able to cope with these situations. The basic paradigm of these studies is to present a pattern of moving dots with a superimposed pattern of scintillating dots. Observers are required to judge some basic aspect of the moving pattern, such as its shape or its direction of movement. The amount of scintillating noise is manipulated to determine the limits of human performance for each attribute to be judged. Based on the results of these experiments we will attempt to develop a computational model that can simulate the performance of human observers on a wide variety of different tasks.