Targetting Number Plates Effectively Using Sparse/Full Templates and Coarse/Fine Template Matching

A coarse-to-fine approach to object location via template matching is described that entails the progression from sparse to full templates, and the progression from large step to small step "marching" of the templates through an image. Our "coarse-tofine" approach may be loosely described as "multi-grid" or "multi-resolution" although it differs in significant detail from previous work. The approach was developed in the context of an assessment of possible fully automated analysis of police camera films of traffic and speed violators. In this paper we describe how our approach has been applied to the location of the border of number plates in selected images. Elsewhere, we shall discuss the application of the "coarse-to-fine" approach to the actual reading of the characters on the number plates so "targetted", allowing for tilt, perspective and scale change. Herein we describe the individual strategies that have been implemented and present theoretical speed-up and actual timing data for performance on a UNIX system. The individual results are generally impressive, although gathered only on a very limited set of images. In this assessment project a complete integrated package has not been developed.