Recognizing structured forms using neural networks

Summary form only given, as follows. It is pointed out that the ability to process complex printed forms automatically is of considerable value to an organization using them. In a stream of heterogeneous forms, it is essential to first recognize a given form as an instance of a particular type, so that its contents can be interpreted correctly. Performing this recognition presents some technical problems that may not be apparent initially. These problems were reviewed, and a neural network implementation was devised that uses a novel feature set to perform reliable identification of US Internal Revenue Service forms