A drawing scene perception model based on the human visual system

Conventional drawing recognition systems can process only objects that are segregated from each other and distinctly regularized in terms of geometric properties such as location, size, and orientation. On the other hand, humans can recognize objects easily even in a complex configuration. This paper proposes a human-like drawing perception model which is composed of eye fixation, figure segregation, complex logarithm coordinate transform, transform to frequency domain, pattern matching, and knowledge pattern activation mechanisms. Results of the computer simulation of the model showed that it can move the direction of gaze to objects to be observed, it can segregate overlapped objects, it can recognize objects invariantly against location, size, and orientation, and it can correctly understand the meaning of scenes composed of several objects