Back to results
US 5,671,291 AGranted
Process and device for character recognition, in particular for standardized character type E-13B
AssigneeDassault Automatismes et Telecommunications
InventorsJean-Philippe Ruault, Richard Besson
Life of this patent
Estimating the term…
Abstract
A wave form representative of the character to be recognized is converted digitally. For each set of N sample reference signals initially acquired, the absolute value of the difference between the numerical value of each sample reference signal and the numerical value of each sample signal to be recognized is calculated in order to establish K lists each of N absolute values. From the K lists is selected that for which the N absolute values have an approximately constant value, where the character to be recognized corresponds to the reference character associated with the list thus selected.
Claims10
A method for character recognition comprising the steps of:
- in a learning phase:
- scanning K reference characters for obtaining K waveforms which are respectively representative of the K reference characters;
- sampling the K waveforms at a predetermined sampling frequency for obtaining N sampled signals for each of the K reference characters;
- determining a numerical value corresponding to an amplitude for each of the N sampled signals of each of the K reference characters;
- storing the numerical values corresponding to the respective amplitudes of each of the N sampled signals of each of the K reference characters;
- in a storage phase:
- scanning a character to be recognized for obtaining a waveform which is representative of the character to be recognized;
- sampling the waveform representing the character to be recognized at the predetermined sampling frequency for obtaining N sampled signals of the character to be recognized;
- determining a numerical value corresponding to an amplitude of each of the N sampled signals of the character to be recognized;
- storing the numerical value corresponding to the amplitude of each of the N sampled signals of the character to be recognized;
- in a decoding phase:
- calculating an absolute value of a difference between the numerical value for each of the N sampled signals of the character to be recognized and a corresponding numerical value of each of the N sampled signals of each of the K reference characters, for establishing K lists, each list having N absolute values and being associated with a unique reference character;
- and
- selecting a list from the K lists based on the absolute values of the selected list having an approximately constant value within the selected list, the character to be recognized being determined to be the reference character associated with the selected list, and
- wherein the method further comprises, in the learning phase, the steps of:
- assessing the numerical values of the N sampled signals of each of the K reference character signals;
- and
- isolating from the N assessed numerical values of each reference character at least two successive numerical values which are greater than a predetermined threshold for obtaining a decoding window for each of the K reference characters.
15 more elements
A method for character recognition comprising the steps of:
- in a learning phase:
- scanning K reference characters for obtaining K waveforms which are respectively representative of the K reference characters;
- sampling the K waveforms at a predetermined sampling frequency for obtaining N sampled signals for each of the K reference characters;
- determining a numerical value corresponding to an amplitude for each of the N sampled signals of each of the K reference characters;
- storing the numerical values corresponding to the respective amplitudes of each of the N sampled signals of each of the K reference characters;
- in a storage phase:
- scanning a character to be recognized for obtaining a waveform which is representative of the character to be recognized;
- sampling the waveform representing the character to be recognized at the predetermined sampling frequency for obtaining N sampled signals of the character to be recognized;
- determining a numerical value corresponding to an amplitude of each of the N sampled signals of the character to be recognized;
- storing the numerical value corresponding to the amplitude of each of the N sampled signals of the character to be recognized;
- in a decoding phase:
- calculating an absolute value of a difference between the numerical value for each of the N sampled signals of the character to be recognized and a corresponding numerical value of each of the N sampled signals of each of the K reference characters, for establishing K lists, each list having N absolute values and being associated with a unique reference character;
- and
- selecting a list from the K lists based on the absolute values of the selected list having an approximately constant value within the selected list, the character to be recognized being determined to be the reference character associated with the selected list,
- wherein the step of selecting comprises the steps of:
- determining a particular absolute value for each of the K lists of absolute values having a value which diverges from the absolute value of a first sample of the character to be recognized;
- sorting the sum results by increasing value;
- and
- selecting a character to be recognized corresponding to the reference character associated with the particular sum result, the selected sum result being less than a predetermined constant.
16 more elements
A method for recognizing an unknown character, comprising the steps of:
- (a) generating a sequence of N sampled values for the unknown character, step (a) including:
- (a-1) scanning the unknown character to obtain a waveform,
- (a-2) sampling the waveform at a predetermined sampling frequency, and
- (a-3) determining numerical values for the samples;
- (b) matching the unknown character with one of K reference characters so as to recognize the unknown character, each of the K reference characters being represented by a stored sequence of N sampled values for the respective reference character, step (b) including:
- (b-1) calculating a sequence of N absolute values of differences between the sampled values of the sequence for the unknown character and the sampled values in the stored sequence for a first one of the reference characters, and additionally calculating a sequence of N absolute values of differences between the sampled values of the sequence for the unknown character and the sampled values in the stored sequences for another of the reference characters, and repeating until sequences of absolute values have been calculated for all K reference characters;
- (b-2) after sequences of absolute values have been calculated for all K reference characters, recognizing the unknown character as said first one of the reference characters if the N absolute values in the sequence are approximately equal, and recognizing the unknown character as a second one of the reference characters if said first one of the reference characters is not recognized as the unknown character and if the absolute values in the sequence of N absolute values for said second one of the reference characters are approximately equal;
- and
- (b-3) calculating a resemblance coefficient for each of the reference characters if the absolute values are approximately equal in none of the sequences of N absolute values, the resemblance coefficient being a function of the sum of the absolute values in the sequence of N absolute values for the respective reference character, and using the resemblance coefficients to recognize the unknown character.
6 more elements
Dependencies are inferred from the claim text.