Confidence modeling for verification post-processing for handwriting recognition

Abstract

We apply confidence-scoring techniques to verify the output of a handwriting recognizer. We evaluate a variety of scoring functions, including likelihood ratios and estimated posterior probabilities of correctness, in a postprocessing mode to generate confidence scores at the character or word level. Using the post-processor in conjunction with an HMM-based on-line handwriting recognizer for large-vocabulary word recognition, receiver-operating-characteristic (ROC) curves reveal that our post-processor is able to reject correctly 90% of recognizer errors while only falsely rejecting 33% of correctly-recognized words. For isolated-digit recognition, we achieve a correct rejection rate of 90% while keeping false rejection down to 13%.