| 摘要 |
PURPOSE:To execute the character recognition which is highly accurate and does not segment a character even when a pattern is not continuous by making the value obtained by segment- linking the concentration vector generated from the picture concentration value of a sampling point with a converting matrix into a characteristic value. CONSTITUTION:A characteristic extracting means 54 outputs the address value of the sampling position in a reading frame and the address of the position in the weight mask scope with the position as the center to a picture memory means 51, obtains the value of the corresponding picture and then, the means multiplies the value and the weighting value of the weight mark by the corresponding values, and outputs the adding value of respective multiplying values to a density vector means 55 as the density characteristic value of a sampling position. A segment linking means 57 calculates the sum of products with the density vector and the vector of the converting matrix value from a converting matrix value memory means 56 and outputs the converting characteristic value to a converting characteristic value memory means 58. A distance counting means 60 counts the distance between the converting characteristic value and the converting characteristic value from a dictionary memory means 59 and outputs the minimum distance value and the corresponding category name to a result memory means 61. A control means 53 controls to repeat the processing.
|
