Pattern recognition using stored n-tuple occurence frequencies
First Claim
Patent Images
1. A pattern recognition device comprising:
- (a) input means for producing a pattern of bits;
(b) means storing assignment data defining predetermined groups of said bits;
(c) a template store; and
(d) control means operating(i) in a training mode to record in the template store, for each of the defined groups the frequency with which combinations of bits within that group occur in a plurality of the said patterns and operating(ii) in a recognition mode in response to an unknown pattern provided by said input means, to retrieve for each defined group of bits selected from said unknown pattern, in response to the bits of the group, the frequency if any which is recorded in the template store for that particular combination of bits, to form a similarity measure for that unknown pattern as a function of the retrieved frequencies, and to produce an output indicative of successful or unsuccessful recognition in dependence on whether the similarity measure meets a recognition criterion, and(iii) in the recognition mode, to adjust the stored frequency in the template store for any group in respect of which a frequency is already stored, for a particular bit combination occurring in the unknown pattern which meets the recognition criterion.
1 Assignment
0 Petitions
Accused Products
Abstract
A pattern recognizer uses the method of n-tuples applied to a matrix containing (e.g.) features derived from input speech. A template store records during a training sequence the frequencies of occurrence of combinations of bits with each n-tuple group. During recognition frequencies associated with combinations which occur are used to form a measure of similarity.
29 Citations
46 Claims
-
1. A pattern recognition device comprising:
-
(a) input means for producing a pattern of bits; (b) means storing assignment data defining predetermined groups of said bits; (c) a template store; and (d) control means operating (i) in a training mode to record in the template store, for each of the defined groups the frequency with which combinations of bits within that group occur in a plurality of the said patterns and operating (ii) in a recognition mode in response to an unknown pattern provided by said input means, to retrieve for each defined group of bits selected from said unknown pattern, in response to the bits of the group, the frequency if any which is recorded in the template store for that particular combination of bits, to form a similarity measure for that unknown pattern as a function of the retrieved frequencies, and to produce an output indicative of successful or unsuccessful recognition in dependence on whether the similarity measure meets a recognition criterion, and (iii) in the recognition mode, to adjust the stored frequency in the template store for any group in respect of which a frequency is already stored, for a particular bit combination occurring in the unknown pattern which meets the recognition criterion. - View Dependent Claims (4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 38, 39)
-
-
2. A pattern recognition device comprising:
-
(a) input means for producing a pattern of bits; (b) means storing assignment data defining groups of bits selected from said pattern; (c) a template store; and (d) control means operating (i) in a training mode to record in the template store, for each of the defined groups the frequency with which combinations of bits within that group occur in a plurality of the said pattern and operating (ii) in a recognition mode in response to an unknown pattern provided by said input means, to retrieve for each defined group of bits selected from said unknown pattern, in response to the bits of the group, the frequency if any which is recorded in the template store for that particular combination of bits, to form a similarity measure for that unknown pattern as a function of the retrieved frequencies, and to produce an output indicative of successful or unsuccessful recognition in dependence on whether the similarity measure meets a recognition criterion, and each template store has for each group a respective section comprising a plurality of entries each comprising a location for storage of a particular bit combination and a location for storage of the frequency associated with that combination.
-
-
3. A pattern recognition device comprising:
-
(a) input means for receiving a pattern of bits; (b) means storing assignment data defining groups of bits selected from the pattern; (c) a template store, and (d) control means operating (i) in a training mode to record in the template store, for each of the defined groups data indicating which combinations of bits within that group occur in the said patterns, and operating, (ii) in a recognition mode in response to an unknown pattern provided by said input means, to retrieve for each defined group of bits selected from the pattern, in response to the bits of the group, an indication as to whether that particular combination of bits is recorded in the template store as having occurred to form a similarity measure for that pattern as a function of the retrieved indicators, and to produce an output indicative of successful or unsuccessful recognition in dependence on whether the similarity measure meets a recognition criterion; the template store having for each group a respective section comprising a plurality of entries each comprising a location for storage of a particular bit combination. - View Dependent Claims (22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37)
-
-
40. In a pattern recognition device of the type which detects and compares occurrences of specific n-tuple features between unknown input and known template patterns to identify the known template pattern most likely corresponding to the unknown input pattern, the improvement comprising:
-
means for storing first data representing a multi-bit value and its occurrence frequency for each of plural predetermined n-tuple features in said known template patterns; means for deriving and temporarily storing second data for detected multi-bit values of said n-tuple feature occurrences in said unknown input pattern, said second data corresponding to any of said known template patterns also representing the same multi-bit values of said n-tuple feature occurrences and being weighted by said first data representing the stored frequency of such n-tuple feature occurrences therein; and means for comparing said second data corresponding to different known template patterns to identify the known template pattern most likely corresponding to the unknown input pattern. - View Dependent Claims (41, 42)
-
-
43. A pattern recognition method of the type which detects and compares occurrences of specific n-tuple features between unknown input and known template patterns to identify the known template pattern most likely corresponding to the unknown input pattern, the method comprising:
-
detecting and storing first data representing a multi-bit value and its occurrence frequency for each of plural predetermined n-tuple features in said known template patterns; deriving and temporarily storing second data for detected multi-bit values of said n-tuple feature occurrences in said unknown input pattern, said second data corresponding to any of said known first template patterns also representing the same multi-bit values of said n-tuple feature occurrences and being weighted by said first data representing the stored frequency of such n-tuple feature occurrences therein; and comparing said second data corresponding to different known template patterns to identify the known template pattern most likely corresponding to the unknown input pattern. - View Dependent Claims (44, 45, 46)
-
Specification