Text image processing using word spacing equalization for icr system employing artificial neural network

Abstract

In an intelligent character recognition (ICR) method for recognizing hand-written text images using a long-short term memory (LSTM) recurrent neural network (RNN), text images are segmented into text line images, and the text lines images are pre-processed to normalize the line height and to equalize the word spacings in each text line. Both training images used to train the RNN network and test images containing text to be recognized by the trained RNN network are pre-processed to have identical heights and identical word spacings between words. This method improves character recognition accuracy.

Description

BACKGROUND OF THE INVENTIONField of the Invention[0001]This invention relates to text image processing, and in particular, it relates to a text image processing method used in conjunction with an artificial neural network to perform intelligent character recognition (ICR) to digitalize hand-written text.Description of Related Art[0002]There is significant interest in applying recurrent neural networks (RNN), such as long short-term memory networks (LSTM), in intelligent character recognition (ICR) systems for hand-written text digitalization. In some applications using RNN neural network for character recognition of hand-written text, text images are segmented into text lines and then into individual words, and word images each containing one word are inputted to the RNN network model, both for training the network model and for performing prediction (character recognition).SUMMARY[0003]In some ICR technologies, word segmentation tasks are challenging due to random spaces between wo...

AI Technical Summary

Benefits of technology

[0005]An object of the present invention is to provide a method to pre-process text line images to reduce the effect of variable word spacings on network training and prediction process, thereby improving recognition accuracy for an ICR system that employs an RNN.

Problems solved by technology

In some ICR technologies, word segmentation tasks are challenging due to random spaces between words in a given text line.

For the same reason, training an RNN network using line-level samples, i.e. text line images each containing one line of text, is often unsuccessful as automatic labeling algorithms such as connectionist temporal classification (CTC) often fail to correctly find correct lengths of spaces between words.

Images