| 摘要 |
<p>1,271,705. Character recognition. SCAN-DATA CORP. 15 Oct., 1969 [8 Nov., 1968], No. 60725/69. Heading G4R. A character recognition system includes a shift register 28 for serially storing and shifting a binary quantization of a character pattern sampled within a field on a document, a plurality of sub-feature masks 36 connected to different combinations of shift-register stages lying in a "window", and a plurality of feature detectors 42 responsive to different combinations of the sub-feature masks and feeding a character decoder 50 via storage means 44. The feature detectors 42 are enabled by appropriate horizontal and vertical position signals from horizontal and vertical counters in horizontal and vertical analysers 38, 40. The character identity from decoders 50 is encoded 62 to feed a central processor 64. A horizontal gap detector 32 uses a mask responsive to the shift register 28 to detect the gap between characters, and in the horizontal analyser 38 this or detection at 42 of the right-side features of a character (the scanning pattern is a rightwards-progressing vertical raster) stops horizontal clock pulses (one per vertical scan line) incrementing the horizontal counter and notifies the processor 64 of the trailing edge of the character so that it will examine the encoder 62 output. The processor resets the horizontal counter for the next character. A count of one indicates the leading edge of the character to the processor. If a count of 32 is reached, the processor causes rescan. A predetermined number of such rescans causes the processor to generate a confusion character to indicate an unreadable character to the operator. The vertical counter is reset by each horizontal clock pulse and incremented by the shift pulses fed to shift register 28. A vertical data column unit 34 receives the contents of one stage of the shift register 28 to produce a horizontal shadow projection of the character in a recirculating shift register, the length of the projection being evaluated by a counter in the vertical analyser 40 to specify whether the character is upper or lower case (where these are identical) and to cause the processor to normalize the vertical scan length if necessary. The position of the scan is also moved up or down in response to line follow logic responding to the recirculating shift register contents so that the scanning adjusts to any deviation of the line of print up or down. The character decoder 50 also receives indications from processor-loaded registers 56, 58 of characteristics of the font being recognized (e.g. serif or sanserif) and font identity, these registers being loaded according to indications from font recognition logic 46 responsive to the feature storage means 44. A majority register 52, loaded by the processor, specifies to the character decoder 50, the number of required features which may be missing without disqualifying a possible character identity. A processor-loaded register 54 inhibits any identification not required, e.g. if only upper case characters on a document are to be read, identities of others are inhibited at the encoder 62. The character decoder 50 may undesirably indicate two or more identities simultaneously (i.e. energize two or more of its output lines simultaneously). A processor-loaded register 60 enables the processor to inhibit all the lines in turn so that the identities are provided by the encoder 62 one at a time. Only a portion (window) of the shift register 28 feeds the rest of the circuitry as above so that the features are detected in turn.</p> |
