Large vocabulary recognition of on-line handwritten cursive words

Abstract

This paper presents a writer independent system for large vocabulary recognition of on-line handwritten cursive words. The system rst uses a ltering module, based on simple letter features, to quickly reduce a large reference dictionary (lexicon) to a more manageable size; the reduced lexicon is subsequently fed to a recognition module. The recognition module uses a temporal representation of the input, instead of a static 2-dimensional image, thereby preserving the sequential nature of the data and enabling the use of a Time-Delay Neural Network (TDNN); such networks havee been previously successful in the continuous speech recognition domain. Explicit segmentation of the input words into characters is avoided by sequentially presenting the input word representation to the neural network-based recognizer. The outputs of the recognition module are collected and converted into a string of characters that is matched against the reduced lexicon using an extended Damerau-Levenshtein function. Trained on 2,443 unconstrained word images (11k characters) from 55 writers and using a 21k lexicon we reached a 97.9% and 82.4% top-5 word recognition rate on a writer-dependent and writerindependent test respectively. Keywords| Handwritten text recognition, on-line cursive handwriting recognition, neural network applications, penbased systems.

An integration of online and pseudo-online information for cursive word recognition

“…Currently, online recognition suffers from several weaknesses that involve sensitivity to stroke order, stroke number, and stroke characteristics variations ( [3]). As similar shapes might be produced by different sets of strokes, two instances of the same letter or word may resemble each other in the image domain though they are associated with diverse online signals.…”

Talking and writing—how natural in human–machine interaction?

“…Results have been encouraging: a word recognition rate of 93.4% has been attained with a 713 word lexicon compared to 49.8% in a post-processing system. Another technique has focused on initial "ltering, based on simple letter features, to reduce the size of the lexicon that is being fed into the recognition process (Seni, Srihari & Nasrabadi, 1996).…”

Online handwriting recognition: the NPen++ recognizer

“…While several preprocessing steps in NPen++ are modeled on existing online handwriting recognition systems [4,7,[20][21][22], some steps, such as context maps, are unique in NPen++. Figure 1 gives an overview of the preprocessing steps of NPen++.…”

“…Computing baselines is a main technique in handwriting recognition and is implemented in several online as well as offline (OCR) systems (e.g., [4,22]). Baselines are utilized for many reasons, e.g., normalizing sizes, correcting rotations, or deriving features.…”