Reducing and controlling sizes of model-based recognizers
AssigneeMicrosoft Corporation
InventorsChristopher Meek, Bo Thiesson, John R. Bennett
Life of this patent
Estimating the term…
Abstract
A technique for reducing a model database for use with handwriting recognizers. The model database is processed with a tuning set to identify a set of models that would result in the greatest character recognition accuracy. If further model database reduction is desired, the technique iteratively identifies smaller models that have the least adverse effect on the error rate. The technique continues identifying smaller models until a desired model database size has been achieved.
Claims36
A method of reducing a model database for use with a handwriting recognizer comprising the steps of:
- (a) processing a plurality of models to form a model set that has a higher accuracy for character recognition, the model set having a first size;
- and
- (b) replacing at least one model from the model set with another model to form a revised model set, the revised model set having a second size smaller than the first size.
A system for reducing a model database comprising in combination:
- (a) a tuning set;
- (b) a model database;
- and
- (c) a model processor having computer-executable instructions for performing the steps of:
- (i) processing a plurality of models to form a model set that has a higher accuracy for character recognition, the model set having a first size;
- and (ii) replacing at least one model from the model set with another model to form a revised model set, the revised model set having a second size smaller than the first size.
A method of reducing a model database for use with a handwriting recognizer comprising the steps of:
- (a) selecting at least one active model for a character;
- (b) determining a change in an error rate if the active model was changed to an alternate model;
- (c) making the alternate model the active model if the error rate is reduced;
- (d) repeating the steps of determining and making for each character to be recognized until an error rate is minimized, thereby forming a model set having a first size;
- (e) replacing at least one model from the model set with another model to form a revised model set, the revised model set having a second size smaller than the first size;
- and
- (f) repeating the step of replacing until the second size of the revised model set reaches a desired size.
A method of optimizing a model database for use with a handwriting recognizer comprising the steps of:
- (a) selecting at least one active model for a character;
- (b) determining a change in an error rate if the active model was changed to an alternate model;
- (c) making the alternate model the active model if the error rate is reduced;
- and
- (d) repeating the steps of determining and making for each character to be recognized until an error rate is minimized.
Dependencies are inferred from the claim text.