Patent EP0063765A2 successfully refreshed from Google Patents.
EP0063765A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Pattern recognition apparatus and method
Assignee: Toshiba Corp
•
Filing: 1982-04-16
•
Granted: —
•
Expires: 2002-04-16
Abstract
Abstract The invention provides a pattern recognition apparatus and method by which an unknown input pattern is divided into a plurality of sub-area patterns matched with a plurality of standard partial characteristic patterns fo form an input pattern similarity matrix. The matrix is then matched with a plurality of reference similarity matrices of standard reference patterns, whereby the unknown pattern is recognized as one of standard reference patterns which gives the best matching. In the apparatus, digital signals corresponding to the sub-area patterns are supplied to a first similarity calculator (15) to which the partial characteristic patterns are supplied from a first dictionary memory (16). The calculator (15) calculates the input pattern similarity matrix. A second similarity calculator (19) calculates similarities between the input pattern similarity matrix and each of reference similarity matrices read out from a second dictionary memory (18). A comparator (20) identifies one reference pattern.
Inventors
- 👤 Maeda Kenichi
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-09 19:55
CPC Classifications
G; G06; G06V; G06V10/00; G06V10/70; G06V10/77; G06V10/772; G06V10/20
Claims Analysis
Total Claims: 32
Claim 1
Independent Claim
243 words
A pattern recognition apparatus for identifying an unknown input pattern as the one of a number of reference patterns to which the input pattern is most similar, the input pattern and reference patterns each being composed of a plurality of sub-area patterns (f), said apparatus comprising input means for converting the input pattern into digital electrical signals, characterized in that said apparatus further comprises first dictionary memory means (16) for storing a plurality of partial characteristic patterns (u); first similarity calculating means (15), connected to said input means and said first dictionary memory means (16), for calculating an input pattern similarity matrix (F) between the sub-area patterns (f) of the input pattern and the partial characteristic patterns (u); second dictionary memory means (18) for storing a number of reference similarity matrices (Φ l ) corresponding to the number of reference patterns, each of the reference similarity matrices (Φ l ) being calculated between the sub-area pat- . terns (f) of one of the reference patterns and the partial characteristic patterns (u); second similarity calculating means (19), connected to said first similarity calculating means (15) and said second dictionary memory means (18), for calculating similarities (S) between the input pattern similarity matrix (F) and each of the reference similarity matrices (Φ l ) ; and comparison means (20), connected to said second similarity calculating means (19), for identifying one reference pattern by comparing the similarities (S) calculated by said second similarity calculating means (19).
Claim 2
Dependent Claim
243 words
A pattern recognition apparatus for identifying an unknown input pattern as the one of a number of reference patterns to which the input pattern is most similar, the input pattern and reference patterns each being composed of a plurality of sub-area patterns (f), said apparatus comprising input means for converting the input pattern into digital electrical signals, characterized in that said apparatus further comprises first dictionary memory means (16) for storing a plurality of partial characteristic patterns (u); first similarity calculating means (15), connected to said input means and said first dictionary memory means (16), for calculating an input pattern similarity matrix (F) between the sub-area patterns (f) of the input pattern and the partial characteristic patterns (u); second dictionary memory means (18) for storing a number of reference similarity matrices (Φ l ) corresponding to the number of reference patterns, each of the reference similarity matrices (Φ l ) being calculated between the sub-area pat- . terns (f) of one of the reference patterns and the partial characteristic patterns (u); second similarity calculating means (19), connected to said first similarity calculating means (15) and said second dictionary memory means (18), for calculating similarities (S) between the input pattern similarity matrix (F) and each of the reference similarity matrices (Φ l ) ; and comparison means (20), connected to said second similarity calculating means (19), for identifying one reference pattern by comparing the similarities (S) calculated by said second similarity calculating means (19).
Claim 3
Dependent Claim
33 words
A pattern recognition apparatus as claimed in claim 1, characterized in that the input pattern, reference patterns, and partial characteristic patterns (µ) are visual patterns composed of picture elements in a plane (11).
Claim 4
Dependent Claim
25 words
A pattern recognition apparatus as claimed in claim 2, characterized in that the digital electrical signals represent the darkness of each of the picture elements.
Claim 5
Dependent Claim
33 words
A pattern recognition apparatus as claimed in claim 3, characterized in that said input means comprises converting means (12, 13) for optically scanning and converting the input pattern into the digital electrical signals.
Claim 6
Dependent Claim
40 words
A pattern recognition apparatus as claimed in claim 4, characterized in that said converting means comprises a rester scanning means (12) for scanning the input pattern and an analog-to-digital converter (13) to convert the input pattern into the digital signals.
Claim 7
Dependent Claim
31 words
A pattern recognition apparatus as claimed in claim 1, characterized in that the input pattern, reference patterns, and partial characteristic patterns are sound patterns composed of areas on a time-frequency plane.
Claim 8
Dependent Claim
76 words
A pattern recognition apparatus as claimed in claim 1, 2, 4 or 6, characterized in that said apparatus further comprises first buffer memory means (14) connected between said input means and said first similarity calculating means (15) for storing the sub-area patterns (f) of the input pattern, and second buffer memory means (17) connected between said first similarity calculating means (15) and said second similarity calculating means (19) for storing the input pattern similarity matrix (F).
Claim 9
Dependent Claim
25 words
A pattern recognition apparatus as claimed in claim 6, characterized in that the digital electrical signals represent the sound amplitude of each of the areas.
Claim 10
Dependent Claim
91 words
A pattern recognition apparatus as claimed in claim 8, characterized in that said input means comprises transducer means (51) for converting the sound patterns into analog electrical signals; frequency division means (53) connected to said transducer means (51) for separating the analog electrical signals into a plurality of frequency bands, and digitizing means (54,-55) connected to said frequency division means (53) for separating the analog electrical signals into a plurality of time segments and producing a digital signal corresponding to the sound amplitude in each time segment in each frequency band.
Claim 11
Dependent Claim
125 words
A pattern recognition apparatus as claimed in claim 9, characterized in that said transducer means comprises a microphone (51) connected to an amplifier (52); said frequency division means (53) comprises a plurality of band pass filters (53-1, 53-2, ..., 53-16) connected to the output of said amplifier (52), said plurality of band pass filters (53-1, 53-2, ..., 53-16) corresponding -to the plurality of frequency bands; and said digitizing means comprises, for each of said band pass filters (53-1, 53-2, ..., 53-16), a detector (54) connected to said band pass 'filter (53) to detect the envelope of the analog signal, and an analog-to-digital converter (55) connected to said detector (54) to produce a digital electrical signal corresponding to the amplitude of the envelope at predetermined intervals.
Claim 12
Dependent Claim
20 words
A pattern recognition apparatus as claimed in claim 6 or 8, characterized in that the partial characteristic patterns comprise phonemes.
Claim 13
Dependent Claim
33 words
A pattern recognition apparatus as claimed in claim 2, characterized in that the partial characteristic patterns (u) comprise at least one end point, at least one intersection point, and at least one loop.
Claim 14
Dependent Claim
17 words
A pattern recognition apparatus as claimed in claim 12, characterized in that the sub-area patterns (f) overlap.
Claim 15
Dependent Claim
72 words
A pattern recognition apparatus as claimed in claim 1, characterized in that said apparatus further comprises blurring means (22), connected between said first similarity calculating means (15) and said second similarity calculating means (l9),.for calculating a blurred input matrix (F') from the input pattern similarity matrix (F), said second similarity calculating means (19) calculating similarities (S) between the blurred input matrix (F') and each of the reference similarity matrices (Φ l ).
Claim 16
Dependent Claim
59 words
A pattern recognition apparatus as claimed in claim 14, characterized in that said blurring means (22) includes a mask pattern memory (23) to store a mask pattern having weighting coefficients, and a multiplier accumulator circuit (24) which is connected to said mask pattern memory (23) and said first similarity calculating means (15) to calculate the blurred input matrix (F').
Claim 17
Dependent Claim
21 words
A pattern recognition apparatus as claimed in claim 15, characterized in that the weighting coefficients are distributed in a Gaussian distribution.
Claim 18
Dependent Claim
38 words
A pattern recognition apparatus as claimed in claim 1, characterized in that said comparison means (20) identifies the one reference pattern corresponding to the largest value of the similarities (S) calculated by said second similarity calculating means (19).
Claim 19
Dependent Claim
38 words
A pattern recognition apparatus as claimed in claim 1, characterized in that said first and second similarity calculating means (15, 19) calculate the similarity S between any two vectors f and f o in accordance with the formula:
Claim 20
Dependent Claim
53 words
A pattern recognition apparatus as claimed in calim 18, characterized in that said first similarity calculating means (15) calculates the input pattern similarity matrix (F) by calculating similarities S between vectors f representing the sub-area patterns of the input pattern and vectors µ representing the partial characteristic patterns, in accordance with the formula:
Claim 21
Dependent Claim
48 words
A pattern recognition apparatus as claimed in claim 18 or 19, characterized in that said second similarity calculating means (19) calculates the similarities S between a vector representing the input pattern similarity matrix F and vectors representing the reference similarity matrices D l in accordance with the formula:
Claim 22
Dependent Claim
237 words
A pattern recognition-apparatus as claimed in claim 1, characterized in that said first and second similarity calculating means are similarity calculating circuits (15, 19) for calculating the similarity between a first vector and a second vector, each of said similarity calculating circuits (15, 19) comprising a first buffer memory (32) to store the first vector, a second buffer memory (34) to store the second vector, a first multiplier accumulator (35) connected to said first buffer memeory (32) and said second buffer memory (34) to calculate the scalar product of the two vectors, a second multiplier accumulator (36) connected to said first buffer memory (32) to calculate the square of the norm of the first vector, a third multiplier accumulator (37) connected to said second buffer memory (34) to calculate the square of the norm of the second vector, a first multiplier (38) connected to said first multiplier accumulator (35) to calculate the square of the scalar product of the two vectors, a second multiplier (39) connected to said second multiplier accumulator (36) and said third multiplier accumulator (37) to calculate the product of the squares of the norms of the first and second vectors, and a divider (40) connected to said first multiplier (38) and said second multiplier (39) to calculate the quotient of the square of the scalar product of the two vectors and the product of the squares of the norms of the two vectors.
Claim 23
Dependent Claim
36 words
A.pattern recognition apparatus as claimed in claim 21, characterized in that said first multiplier accumulator (35) calculates the scalar product (f, f o ) of the two vectors f and f o according to the formula:
Claim 24
Dependent Claim
38 words
A pattern recognition apparatus as claimed in claim 21 or 22, characterized in that said second and third multiplier accumulators (36, 37) calculate the squares f 2 of the norms of vectors f in accordance with the formula:
Claim 25
Independent Claim
137 words
A method of identifying an unknown input pattern as the one of a number of reference patterns to which the input pattern is most similar, characterized in that said method comprises the steps of storing a plurality of partial characteristic patterns (p), dividing the input pattern and reference patterns into a plurality of sub-area patterns (f), calculating an input pattern similarity matrix (F) between the sub-area patterns (f) of the input pattern and the plurality of partial characteristic patterns (µ), calculating a plurality of reference similarity matrices (Φ l ) between the sub-area patterns (f) of each reference pattern and the partial characteristic patterns (µ), calculating similarities (S) between the input pattern similarity matrix (F) and each of the reference similarity matrices (Φ l ), and comparing the similarities (S) in order to identify the input pattern.
Claim 26
Independent Claim
137 words
A method of identifying an unknown input pattern as the one of a number of reference patterns to which the input pattern is most similar, characterized in that said method comprises the steps of storing a plurality of partial characteristic patterns (p), dividing the input pattern and reference patterns into a plurality of sub-area patterns (f), calculating an input pattern similarity matrix (F) between the sub-area patterns (f) of the input pattern and the plurality of partial characteristic patterns (µ), calculating a plurality of reference similarity matrices (Φ l ) between the sub-area patterns (f) of each reference pattern and the partial characteristic patterns (µ), calculating similarities (S) between the input pattern similarity matrix (F) and each of the reference similarity matrices (Φ l ), and comparing the similarities (S) in order to identify the input pattern.
Claim 27
Dependent Claim
24 words
The method of claim 24, characterized by further comprising the step of blurring the input pattern similarity matrix (F) before calculating the similarities (S).
Claim 28
Dependent Claim
25 words
The method of claim 25, characterized by further comprising the step of blurring the reference similarity matrices (Φ l ) before calculating the similarities (S).
Claim 29
Dependent Claim
36 words
The method of claim 24, characterized in that the input pattern is a visual pattern and the method further comprises the step of optically scanning the input pattern before calculating the input pattern similarity matrix (F).
Claim 30
Dependent Claim
26 words
The method of claim 24, characterized in that the input pattern is identified as the reference pattern corresponding to the largest value of the similarities (S).
Claim 31
Dependent Claim
21 words
The method of claim 24, characterized in that the input pattern, reference patterns, and partial characteristic patterns (µ) are sound patterns.
Claim 32
Dependent Claim
67 words
The method of claim 29, characterized in that the step of dividing the input pattern and reference patterns into a plurality of sub-area patterns (f) comprises the steps of dividing each pattern into a plurality of frequency bands, dividing each pattern into a plurality of time segments, and generating digital signals representing the sound amplitude of the input pattern within each frequency band during each time segment.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.