Dept. of Computer Science and Engineering
Oregon Graduate Center
David Marr has proposed a computational model of early vision. This model uses the current understanding of the physiology of the human visual system as an intuitive basis for the discovery of the algorithms necessary for machine early vision [Marr82]. This thesis will describe how to implement Marr's model by analysis of each computational task comprising early vision. The first task is sampling the visual world. This computation builds a discrete two-dimensional intensity array. Next, digital filtering techniques are used to construct symbolic primitives which form an intermediate representation, from which the raw primal sketch is built [Marr76]. These primitives are the zero-crossings of the second directional derivative taken in all orientations and at a number of different spatial scales throughout the intensity array. Marr et al. have argued that images encoded with zero-crossing primitives contain sufficient symbolic information to reconstruct the original visual image, and that these primitive symbols are formed into tokens for manipulation by higher-order vision algorithms [CrMP80].
Lulich, Daniel P., "Zero-crossings symbolic vision primitives emulating physiologic encoding schemes" (1985). Scholar Archive. 90.