GB1255654A - Improvements in or relating to character recognition systems - Google Patents

Abstract

1,255,654. Character recognition. MULLARD Ltd. 28 March, 1969 [26 July, 1968], No. 35873/68. Heading G4R. Apparatus for processing a thin digitized character so as to effectively reconstruct the character by joining the points into continuous lines uses alternately (a) a forward search with a biased direction to find successive points of a line, stopping at a line end or on detection of a loop, (b) a radial search from the points of the line found to locate further points which are then joined to the line. The invention is applied in a character recognition system to a digitized character, thinned to one bit thickness as in Specification 1,153,703 or 1,255,653. Successive points are joined by a join from one matrix point to the next along a row (2 possible directions), column (2 directions) or diagonal (4 directions), or by a sequence of such joins. The forward search conducted from a given point on the line is biased in that one of the 8 directions taken by the (last) join leading to the given point and consists of considering matrix points ahead and to the sides of the given point in turn giving priority to those nearest to the given point and closest to the ahead direction until a black point is found. This point is marked. The co-ordinates of the character (black) points are stored in a list of character points in a computer store. The list is initially searched for an unmarked point (the sequence can be restarted if it terminates prematurely), from which a radial search pattern is conducted to locate another black point which is marked and a forward search in the (last) join direction performed, this being repeated to trace out the character line until a forward search gives no result (line end) or arrives at an already marked point (loop). Then a radial search is performed, the first point in the search pattern being considered starting from each marked point (these having been placed in an output list) in turn, then the second point in the search pattern starting from each marked point in turn, and so on until a black point is found, this being marked. The radial search pattern is a (digitized) outwardly progressing spiral. In the above process of alternating searches, which continues until all the character points are marked (" used "), line-end points are marked once, points at line junctions are marked three or more times, other points are marked twice, and small loops and small line segments are ignored. Line ends and junctions are detected on the basis of number of marks, in the output list. Loops (of sufficient size) were marked in a results list as they were detected. The searches used respective search lists of increments (to point co-ordinates). During operation, lists of ends and junctions are formed, and the mean directions of lines joining them are stored, together with whether the lines are essentially straight or curved.