| 摘要 |
PURPOSE:To improve the recognition factor, by securing matching of input patterns with a free end point and therefore deleting the boundary part of patterns generated consecutively to perform matching. CONSTITUTION:A feature extracting means 1 converts an input signal into a series of feature vector, and a word standard pattern storage means 2 which stores a standard pattern formed by the feature vector series. Then a standard pattern train having the minimum distance of cumulative collation between the standard pattern and an input pattern is obtained by a cumulative distance calculator 4 in a dynamic planning method. In this case, the distance between the feature vector of the standard pattern and the input pattern of a nearby frame to the frame of the input pattern corresponding to the initiation or termination frame of the standard pattern to be collated is replaced with fixed estimated value to obtain a cumulative collation distance. |
