Confidence Measures for an Address Reading System
ICDAR '03 Proceedings of the Seventh International Conference on Document Analysis and Recognition - Volume 1
| Hi-index | 0.00 |
Abstract: In this paper a scheme for handwriting adaptation for post offices is described to improve recognition performance of German addresses. The recognition system is based on a tied-mixture Hidden Markov Model (HMM), whose parameters are updated using the expectation maximization (EM) technique, the maximum likelihood linear regression (MLLR) algorithm and a new discriminative adaptation technique, the scaled likelihood linear regression (SLLR). Contrary to the usual approach of adapting a writer-independent system to a specific writer, we propose here to adapt the system to the writer-independent data of a specific post office. The resulting system for each post office yields up to 16% lower word recognition errors.