Concurrent two-stage multi-network optical character recognition system
First Claim
1. A character recognition system, comprising:
- a soft pre-classifier responsive to a representation of an input character to provide a first output signal comprising a coarse vector estimating probabilities of the input character belonging to each one of multiple predefined groups of recognized target characters;
for each particular group of target characters, a specialized neural network to receive the input character and provide a second output signal comprising a fine vector estimating probabilities of the input character representing each one of multiple target characters in the particular group of target characters;
a multiplier, coupled to the soft pre-classifier and each of the specialized neural networks, to weight each specialized neural network'"'"'s second output signal in proportion to the first output signal to provide multiplier output lists, each multiplier list including a list of target characters each associated with a confidence value indicating a probability that the associated target character correctly represents the input character; and
a candidate selector, coupled to the multiplier, to compile and rank contents of the multiplier output lists and provide an aggregate list of target characters and associated confidence values.
1 Assignment
0 Petitions
Accused Products
Abstract
A multi-stage multi-network character recognition system decomposes the estimation of a posteriori probabilities into coarse-to-fine stages. Classification is then based on the estimated a posteriori probabilities. This classification process is especially suitable for the tasks that involve a large number of categories. The multi-network system is implemented in two stages: a soft pre-classifier and a bank of multiple specialized networks. The pre-classifier performs coarse evaluation of the input character, developing different probabilities that the input character falls into different predefined character groups. The bank of specialized networks, each corresponding to a single group of characters, performs fine evaluation of the input character, where each develops different probabilities that the input character represents each character in that specialized network'"'"'s respective predefined character group. A network selector is employed to increase the system'"'"'s efficiency by selectively invoking certain specialized networks selected, using a combination of prior external information and outputs of the pre-classifier. Relative to known single network or one-stage multiple network recognition systems, the invention provides improved recognition, accuracy, confidence measure, speed, and flexibility.
95 Citations
43 Claims
-
1. A character recognition system, comprising:
-
a soft pre-classifier responsive to a representation of an input character to provide a first output signal comprising a coarse vector estimating probabilities of the input character belonging to each one of multiple predefined groups of recognized target characters; for each particular group of target characters, a specialized neural network to receive the input character and provide a second output signal comprising a fine vector estimating probabilities of the input character representing each one of multiple target characters in the particular group of target characters; a multiplier, coupled to the soft pre-classifier and each of the specialized neural networks, to weight each specialized neural network'"'"'s second output signal in proportion to the first output signal to provide multiplier output lists, each multiplier list including a list of target characters each associated with a confidence value indicating a probability that the associated target character correctly represents the input character; and a candidate selector, coupled to the multiplier, to compile and rank contents of the multiplier output lists and provide an aggregate list of target characters and associated confidence values. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27)
-
-
28. A method for character recognition, comprising the steps of:
-
responsive to a representation of an input character, providing a first output signal comprising a coarse vector estimating probabilities of the input character belonging to each one of multiple groups of characters; responsive to the input character, providing multiple second output signals each corresponding to a particular group of characters and comprising a fine vector estimating probabilities of the input character representing each one of multiple characters in the particular group of characters; weighting each second output signal in proportion to the first output signal to provide multiplier output lists each comprising an estimate of individual probabilities that the input character corresponds to each of the characters of a different specialized neural network; and assembling and ranking contents of the multiplier output lists to provide individual ranked estimates according to a selected criteria. - View Dependent Claims (29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43)
-
Specification