Abstract
We propose a feature-segmentation model for the representation of a single pattern in short-term visual memory(STVM). The structure of the model is that three processes interact to determine a memory sensitivity(d'): aquisition, memory, and decision. The basic assumptions of the model can be expressed such that(1)Forgetting occures because the representation of each convex part is individually distorted with memory noise, (2)The amount of memory noise with each convex part is fixed ragardless of pattern complexity, (3)The probability that memory noise occurs is nearly zero if the presented figure is within memory capacity, (4)Memory noise increases as a linear function of the retention interval. Simulation showed that the data of Sakai & Inui(1999a, 1999b, 1999c)were quite compatible with the present model. It is suggested that segmented convex parts are stored in STVM with a parallel rehearsal.
