Garbled Text String Recognition with a Spatio-Temporal Pattern Recognition NeuralNetwork (Verminkte Tekst String Herkenning met een Spatio-Temporeel Patroon Herkennings Neuraal Netwerk)

摘要

The purpose of this project is to show that neural networks can be used torecognize garbled words and/or parts of sentences in a real-world application. TNO-FEL has studied and built an application that recognizes the names of ships. These names may be garbled by transmission or typing errors, and synonyms or corruptions may be used. The SPR network emphasizes the character-sequence relationships within words. SPR is proof against missing, extra or interchanged (pairs of) characters. A learning strategy was developed and implemented. Several measurements were performed. Keywords: Netherlands. (Author) (kr)