Techniques for dynamic interactive sketch recognition

Tools

Baker, N., 1990. Techniques for dynamic interactive sketch recognition. MPhil, Nottingham Trent University.

Preview

Text
10183546.pdf - Published version
Download (16MB) | Preview

Official URL: http://gateway.proquest.com/openurl?url_ver=Z39.88...

Abstract

The objective of this work is to develop pattern recognition techniques which will render hand-drawn sketches to draughtsman-like quality, suitable for real-time operation in an interactive human/computer interface. The intended use is as a means of entering graphical information in the office environment.

The recognition process has been rationalised into discreet stages with the intention pf a) reducing complexity, by simplifying the interface between different recognition functions, and b) providing a suite of recognition functions, suitable for use in a variety of sketch recognisers, of varying sophistication.

Drawing takes place with a pen on a digitising tablet. The position of the pen on the tablet is sampled at frequent time intervals, thus the pen path is digitised into a stream of points. Recognition stages envisaged are:-

Input Filtering (chapter 5)

Stroke Segmentation (chapter 4)

Curve Fitting (chapter 6)

Stroke Analysis (chapter 7)

Shape Matching, and Hierarchical Analysis (which are outside the scope of this work).

Input Filtering filters out sampled points which are determined to be too close to the previous sample to be useful to the stroke segmentation process.

Stroke Segmentation distinguishes straight from curved regions in a pen stroke. The angular variation in the pen path is monitored using a scan-along algorithm to classify regions. As soon as there is a break in classification Curve Fitting renders the recognised region of the stroke with a geometric line or curve.

Stroke Analysis compares a newly fitted line to previous lines in the sketch in an attempt to merge it with previous lines. This is in response to the observed drawing habit, that a user sometimes represents a single line by drawing it in several fragments. Detection at this stage, using heuristics, simplifies the task of Shape Matching.

Shape Matching takes a newly fitted, or merged line, and searches the previous lines in the sketch, to recognise elementary shapes from a given set of templates.

Hierarchical Analysis is a generic term for further stages of recognition, probably employing spatial relations between elementary shapes and lines, and grammars.

The stages of recognition addressed in this thesis are Input Filtering, Stroke Segmentation, and Stroke Analysis. Shape matching and hierarchical analysis, beyond the scope of study, are higher levels of recognition which can benefit from the preprocessing studied here.