An automated numeral reading system is developed and applied to VISA® credit card application forms. The numeric fields were extracted from paper form which contains handprinted personal social security numbers and telephone numbers. The proposed reading system consists of three modules. The first module is a preprocessing module which deals with the tasks of raw form scanning and noise cleaning. The locating module detects the numeric field locations for extraction purpose using a search algorithm. The recognition module uses a hybrid neural model that consists of both unsupervised and supervised learning networks. In this paper, we present experimental results which demonstrate the ability to read handwritten numeric fields. Experiments were performed on a test data set from the CCL/ITRI Database which consists of over 90,390 handwritten numeric digits. In addition, we investigated reliability of the reading system for industry application.
All Science Journal Classification (ASJC) codes
- Computer Science(all)