Application of Deformable Template Matching to Symbol Recognition in Hand-written Architectural Drawings

We propose to use deformable template matching as a new approach to recognise characters and lineal symbols in hand-written line drawings, instead of traditional methods based on vectorization and feature extraction. Bayesian for-mulation of the deformable template matching allows com-bining fidelity to the ideal shape of the symbol with maxi-mum flexibility to get the best fit to the input image. Lineal nature of symbols can be exploited to define a suitable rep-resentation of models and the set of deformations to be ap-plied to them. Matching, however, is done over the original binary image to avoid losing relevant features during vec-torization. We have applied this method to hand-written ar-chitectural drawings and experimental results demonstrate that symbols with high distortions from ideal shape can be accurately identified. 1