Similarity calculator comprising a buffer for a single input pattern feature vector to be pattern matched with reference patterns

Claim 1

A similarity calculator for calculating a set of similarity measures S(A(u, m), B c )'s between every one of a set of fragmentary patterns A(u, m)'s of an input pattern A and every one of first through N-th reference patterns B c (c=1, 2, . . . , N), said input pattern A being given by a time sequence of first through I-th input pattern feature vectors a i (i=1, 2, . . . , I), said set of fragmentary patterns A(u, m)'s being defined by an end point m (m being representative of each of i) and a plurality of start points u's predetermined relative to the end point m as those parts of said time sequence which consist of (u+1)-th through m-th input pattern feature vectors a u+1 to a m , an n-th reference pattern B n (n being representative of each of c) being given by a time sequence of first through J n -th reference pattern feature vectors b j n (j n =1, 2, . . . , J n ), said similarity calculator comprising: input pattern buffer means for producing said input pattern feature vectors a i successively in the ascending order, the m-th input pattern feature vector a m during an m-th relatively long period of time; a reference pattern memory for memorizing the reference pattern feature vectors b j c of the respective reference patterns B c ; control means operatively coupled to said input pattern buffer means for producing a reference pattern specifying signal specifying said reference patterns B c successively in a predetermined order in the m-th relatively long period, the n-th reference pattern B n during an n-th period of first through N-th intermediate poriods of time, a first set of signals specifying in a first fragment of the n-th intermediate period the first through the J n -th reference pattern feature vectors b j n , respectively, an integer specifying signal specifying integers 1 through (J n +r) (r being representative of a predetermined integer) successively in the descending order in a second fragment next succeeding the first fragment in the n-th intermediate period, one integer k (k being representative of each of 1 through [J n +r]) during a k-th period of (J n +r)-th through first relatively short periods of time, and a second set of signals specifying in each relatively short period in the n-th intermediate period those consecutive ones of the first through the J n -th reference pattern feature vectors b j n , respectively, which are prescribed relative to the integer k specified during said each relatively short period; reference pattern output means connected to said reference pattern memory and respective to the reference pattern specifying signal and the first-set signals for making said reference pattern memory produce the first through the J n -th reference pattern feature vectors b j n in the n-th intermediate period; scalar product calculating means connected to said input pattern buffer means and said reference pattern output means for calculating, in the n-th intermediate period of the m-th relatively long period, scalar products (a m ·b j n ) of the m-th input pattern feature vector a m and the respective reference pattern feature vectors b j n produced in the meantime; scalar product buffer means connected to said scalar product calculating means and said control means and comprising a plurality of buffer addresses specifiable by the signals of whichever of the first and the second sets for holding the scalar products (a m ·b j n ) in the buffer addresses specified by the first-set signals and for producing the scalar products (a m ·b j n ) from the buffer addresses specified by the second-set signals; and recurrence value calculating means connected to said scalar product buffer means and responsive to the reference pattern specifying signal, the integer specifying signal, and the second-set signals for carrying out, during the k-th relatively short period in the n-th intermediate period of the m-th relatively long period and for a v-th reference pattern feature vector b v (v being representative of each of j n ) specified in the meanwhile by a v-th one of the second-set signals, calculation of a recurrence value g k (m, v) by the use of the scalar product (a m ·b v ) produced from the buffer address specified by the last-mentioned second-set signal and the recurrence values comprising three recurrence values g k-1 (m-1, v)'s calculated during the (k-1)-th relatively short period in the n-th intermediate period of the (m-1)-th relatively long period and for three consecutive reference pattern feature vectors b v 's predetermined relative to the v-th reference pattern feature vector b v , said recurrence value calculating means providing the similarity measure set S(A(u, m), B c )'s at the end of the m-th relatively long period.