On-line segmentation of freehand sketches by knowledge-based nonlinear thresholding operations

This paper describes an on-line procedure for segmenting freehand sketches into 2D primitives and smooth curves by knowledge-based nonlinear thresholding operations. Drawing position, direction, speed and acceleration information is used for establishing fuzzy knowledge and inferring user's intention. The procedure can be divided into three stages: (1) obtaining corner points based on the fuzzy knowledge and adaptive thresholds, which includes detecting "obtuse" and "acute" corner points, (2) finding smooth join points, during which stage we use both curve linearity and curvature; and (3) classifying the sketches into the corresponding curves according to curve linearity and changes of curve convexity. The method is illustrated and tested with a number of sketches.