A novel multiple-stage approach to the recognition of handwritten words is presented, and its characteristics investigated using a database of words extracted from British cheques. In terms of both recognition performance and pattern throughput the proposed approach offers great flexibility in adapting to various operating constraints, allowing optimisation with respect to differing task requirements. It is demonstrated that this multiple-stage technique can enhance recognition accuracy in comparison with the performance of the component stages. In addition, this multiple-stage approach delivers higher throughput than that achievable by the individual stages working alone. The proposed system is therefore shown to combine high recognition rate with high throughput, a combination well suited to many practical applications.
展开▼