Structural classification and relaxation matching of totally unconstrained handwritten zip-code numbers

摘要

A system for recognizing totally unconstrained handwritten numerals is described. It comprises a feature extractor and two classification algorithms. The feature extractor decomposes the skeleton of a character into geometric primitives containing topological information of the character. These primitives consist of convex polygons and line segments, and features are generated from each primitive. The recognition process contains a fast structural classifier that identifies the majority of the samples, and a robust relaxation algorithm which classifies the rest of the data. The system was trained and tested on real-life handwritten ZIP codes.