Pattern recognition system

Claim 2

A pattern recognition system comprising: pattern-extracting means for converting an input signal into a time sequence A i (i=1, 2, . . . , I) of characteristic patterns; representative reference template memory means for storing representative reference templates B m (m=1, 2, . . . , M) from among a plurality of reference templates B m , said representative reference templates having smallest distances among said reference templates to the others of said reference templates; reference template memory means for storing reference templates B n m (n=1, 2, . . . , N) which are other than said representative reference templates B m (m=1, 2, . . . , M) and have same feature parameters as said representative reference templates; pre-matching computation means for computing a matching between said time sequence A i (i=1, 2, . . . , I) of characteristic patterns and said representative reference templates B m ; pre-matching means for outputting a number j of smallest pre-matched distances, which are given by said pre-matching computation means, together with j corresponding category codes, where j is greater than 2; computing means for computing a matching between said time sequence A i (i=1, 2, . . . , I) of characteristic patterns and only those of said reference templates which correspond to said category codes issued from said pre-matching means among said reference templates stored in said reference template memory means; averaging means for producing an average of a smallest k distance(s) for each group of: (a) distances between said time sequence /A i (i=1, 2, . . . , I) and said reference templates B n m (n=1, 2, . . . , N), and (b) a distance between said time sequence A i (i=1, 2, . . . , I) and said representative reference template B m obtained by said pre-matching computation means, for each of said j category codes obtained from said prematching means; and recognition means for recognizing a category code corresponding to a group which includes a representative reference template B m and reference templates B n m (n=1, 2, . . . , N) that are nearest to said time sequence A i (i=1, 2, . . . , I).

Claim 3

A pattern recognition system comprising: pattern-extracting means for converting an input signal into a time sequence A i (i=1, 2, . . . , I) of characteristic patterns; representative reference template memory means for storing representative reference templates B m (m=1, 2, . . . , M) from among multiple reference templates for each category m (m=1, 2, . . . , M), said representative reference template having smallest distance among said reference templates of a category to the other reference templates of said category; reference template memory means for storing reference templates B n m (n=1, 2, . . . , N) which are other than said representative reference templates B m (m=1, 2, . . . , M), and have same feature parameters as said representative reference templates; pre-matching computation means for computing a matching between said time sequence A i (i=1, 2, . . . , I) of characteristic patterns and said representative reference templates B m (m=1, 2, . . . , M), pre-matching means for comparing a smallest one of a plurality of distances between said time sequence A i (i=1, 2, . . . , I) and said representative reference template B m given as a result of computing by said pre-matching computation means, with a predetermined threshold distance to obtain a comparison result, and when the comparison result is that said smallest distance is smaller than said threshold distance, outputting only a category code corresponding to a representative reference template which gives the smallest distance, and when the comparison result is that said smallest distance is larger than said threshold distance, outputting a smallest j (j≧2) pre-matched distances, which are given by said pre-matched computation means, together with j corresponding category codes; gate means for outputting said category code as a recognition result when the output of said pre-matching means is ony one category code, and outputting a signal to carry out a matching computation between said reference templates stored in said reference template memory means and said time sequence A i when said output of said pre-matching means is a pre-matched distance to multiple category codes; computing means for computing matching said time sequence A i (i=1, 2, . . . , I) of characteristic patterns with only those of said reference templates which correspond to said category codes issued from said pre-matching means, among reference templates stored in said reference template memory means; averaging means for producing an average of a smallest k distance(s) for each group of: (a) distances between said time sequence A i (i=1, 2, . . . , I) and said reference templates B n m (n=1, 2, . . . , N), and (b) a distance between said time sequence A i (i=1, 2, . . . , I) and said representative reference template B m obtained by said pre-matching computing means, for each category code obtained from said pre-matching means; and recognition means for recognizing a category code corresponding to a group, which consists of a representative reference template B m and reference templates B n m (n=1, 2, . . . , N) and is nearest to said time sequence A i (i=1, 2, . . . , I).

Claim 4

A pattern recognition system comprising: pattern-extracting means for converting an input signal into a time sequence A i (i=1, 2, . . . , I) of characteristic patterns; representative reference template memory means for storing representative reference templates B m (m=1, 2, . . . , M) from among multiple reference templates for each category m (m=1, 2, . . . , M), said representative reference template having smallest distance among said reference templates of a category to the other reference templates of said category; reference template memory means for storing reference templates B n m (n=1, 2, . . . , N) which are other than said representative reference templates B m (m=1, 2, . . . , M), and have same feature parameters as said representative reference templates; pre-matching computation means for computing a matching between said time sequence A i (i=1, 2, . . . , I) of characteristic patterns and said representative reference templates B m (m=1, 2, . . . , M), pre-matching means for comparing a smallest one of a plurality of distances between said time sequence A i (i=1, 2, . . . , I) and said representative reference template B m given as a result of computing by said pre-matching computation means, with a predetermined threshold distance to obtain a comparison result, and when the comparison result is that said smallest distance is smaller than said threshold distance, outputting only a category code corresponding to a representative reference template which gives the smallest distance, and when the comparison result is that said smallest distance is larger than said threshold distance, outputting a smallest j (j≧2) pre-matched distances, which are given by said pre-matched computation means, together with j corresponding category codes; gate means for outputting said category code as a recognition result when the output of said pre-matching means is ony one category code, and outputting a signal to carry out a matching computation between said reference templates stored in said reference template memory means and said time sequence A i when said output of said pre-matching means is a pre-matched distance to multiple category codes; computing means for computing matching said time sequence A i (i=1, 2, . . . , I) of characteristic patterns with only those of said reference templates which correspond to said category codes issued from said pre-matching means, among reference templates stored in said reference template memory means; averaging means for producing an average of a smallest k distance(s) for each group of: (a) distances between said time sequence A i (i=1, 2, . . . , I) and said reference templates B n m (n=1, 2, . . . , N), and (b) a distance between said time sequence A i (i=1, 2, . . . , I) and said representative reference template B m obtained by said pre-matching computing means, for each category code obtained from said pre-matching means; and recognition means for recognizing a category code corresponding to a group, which consists of a representative reference template B m and reference templates B n m (n=1, 2, . . . , N) and is nearest to said time sequence A i (i=1, 2, . . . , I).

Claim 5

A pattern recognition system in accordance with claim 1, wherein said pre-matching computation means has same matching computing means as that of said computing means.

Claim 6

A pattern recognition system in accordance with claim 2 wherein said pre-matching computation means has same matching computing means as that of said computing means.