Components of a Line Drawing Interpretation System

Abstract

To date, image interpretation systems have been most successfully employed in industrial inspection tasks, bin picking exercises and for robotic guidance [36], though more recently attention has shifted towards more human-oriented applications such as face recognition and automatic visual surveillance. Most systems operate within structured environments where the data to be assessed is both well-known and minimised; the appearance of objects of interest can often be quite accurately predicted, with each object typically occupying only a small proportion of the image. Line drawings are a means of communication between humans. As they have evolved, drawing styles have emerged which rely heavily on the interpretive power of the human visual system. Hence, although line drawing interpretation might seem more tractable than a ‘natural’ vision problem like, say, the extraction of three-dimensional descriptions of walking people from a video tape of a street scene [37], the task is in fact highly complex.