An improved line continuation model in human visual perception

摘要

An improved line continuation model is presented based upon the concepts of closure property and selective attention in human visual perception. This model can perform the line segregation from a given line picture with or without noise. The original line continuation model, which has confirmed the ability of parallel processing and association, can work well in dealing with natural well-defined thick-line patterns. However, the treatment for noisy or dotted line pictures remains an open problem. This paper continues this topic. To find the correct continuation from a noisy line image, a well-known lateral-interaction process, conducted from a physiological phenomenon called “field of induction” on the retina, is involved in the improved model. The law of inverse square is used to produce an initial (orientational) field which is a diffusion image. Because the simple diffusion of stimulus signals will cause a sharp angle to become a smooth curve, a selective attention mechanism (simulated by a defined orientation force) is constructed to modify the field. The field of induction is iteratively modified by the selective attention algorithm until the attended field is obtained and can be processed by the original model to segregate the individual line patterns from the given line image. Experiments show that the improved model can perform the line segregation well for the dotted and noisy line pictures that the original model cannot deal with.