Perceptual simplification and vectorization of paper-based scribbles

Abstract

This dissertation studies the problem of scribble vectorization, investigating the use of Gabor filtering, sparse pixel tracking and Kalman filtering as a solution to scribble vectorization. Automated simplification and vectorization of scribbles is essential in early-conceptual design allowing the designer to rapidly obtain digital representations of the initial scribbles which may then be used to create 3D virtual CAD models. Since traditional pen-and-paper are still used by designers during early form design, this study investigates algorithms that may be used on paper-based scribbles, allowing designers to retain the preferred drawing medium. The proposed scribble simplification is carried out by first processing the scribble with a Gabor grouping scheme which groups over-traced line strokes into single line groups. The proposed Gabor grouping scheme consists of a quadrature filter bank which groups all over-traced strokes and an inhibiting filter bank which inhibits the quadrature filter response at inter-group gaps present between groups of overstrokes. After this, a tracking algorithm which tracks the simplified scribble sparsely is defined. The tracking algorithm makes use of Kalman filtering to reduce the error in the estimated track points, hence representing the scribbled drawing by smooth, piecewise linear paths.