| 摘要 |
<p>1,171,627. Character recognition. H.M. POSTMASTER GENERAL. 10 Oct., 1967 [7 Oct., 1966; 16 Feb., 1967], Nos. 45045/66 and 7560/67. Heading G4R. In a character recognition machine, the character is stored in a matrix of storage cells which are then scanned from a predetermined one of the cells, examining the cells adjacent to it in a predetermined order, moving to the first storing a portion of the character, examining the cells adjacent to this cell, and so on, whereby the character is expressed as a sequence of movement directions. The character is scanned and copied into a first matrix of storage cells. The size of the character is determined to fix a parameter D. Under computer control the cells of the matrix are considered in turn starting at the top left corner and moving along each row in turn to locate the character. When a point of the character is reached each character branch of length greater than D therefrom is followed in turn, gaps of a size determined by D being filled in and the resulting smoothed character being established in a second storage matrix with progressive clearing of the first matrix. Both matrices are consulted during the process. When all branches from a given character point have been followed the row-by-row search continues. A chain code representing the character as a sequence of 1 out of 8 equally-spaced directions together with node indicators is developed. For each node, an identifying number is assigned and the co-ordinates of the node and the number of branches at the node are specified. The contents of the second matrix are now transferred to the first matrix (since cleared). A starting point on the character (in the first matrix) is obtained as before except that the row-by-row search only extends halfway across the width of the character, and when a character point is found the computer hunts in a 3 x 3 square of cells around it for a node having a higher number of branches than the starting point this node, if found, being taken as the starting point instead. Ceriphs &c. at the starting point are eliminated in this way. The ignoring of branches shorter than D eliminates them elsewhere. The character is followed and transferred to the second matrix with line thinning (except at nodes), the second matrix being consulted before a given branch is transferred to ensure that no branch from the same point in the same direction has already been transferred. A chain code for the character now in the second matrix is derived as before and simplified by eliminating fluctuations from average directions to some extent and then by eliminating all but one of each sequence of identical direction indicators. The reduced chain code is compared with reference codes to identify the character, allowance being made for variants, including alteration of the chain code in a mechanical way (described) to compensate for angular misorientation of the character.</p> |
