-
公开(公告)号:DE1499377A1
公开(公告)日:1969-11-06
申请号:DE1499377
申请日:1965-04-09
Applicant: IBM
Inventor: LEE MALABY DAVEY
Abstract: 1,077,094. Character recognition. INTERNATIONAL BUSINESS MACHINES CORPORATION. March 30, 1965 [April 16, 1964], No. 13341/65. Heading G4R. Deflection voltages for a cathode ray tube document scanner, dependent on time intervals represented by digital inputs, are corrected in accordance with departure of beam movement between two reference marks on the document, from a standard value. The beam positioning of a cathode ray tube used for scanning a document for character recognition purposes is controlled by variable length control words supplied by a digital computer. Each word controls initial positioning and subsequent movement of the beam to read a selected field of the document or for error measurement purposes (see below). In any one of these operations, the characters of the word are supplied in turn by the computer at 11À5 microseconds intervals. The first character specifies the operation and is followed by a blank second character to allow time for resetting within the system. The remaining characters may contain bits which start and stop various " format " integrators used for preparing and storing deflection voltages, there being one bit position corresponding to each of these "format" " integrators, an integrator being stopped when a character arrives with its bit present. Two or more bits may be present in the same character (thus stopping two or more integrators at the same time) but one particular combination of present bits is used (only) to start all the " format " integrators. The " format " integrators are: horizontal fine and coarse, vertical fine and coarse, and end-offield fine and coarse. When on, each coarse integrator increases its stored voltage by enough to deflect the CRT beam by 100 mils, per character time (11À5 microseconds). The corresponding figure for each fine integrator is 10 mils. When the previous operation has been completed, the voltages in the " format " integrators are transferred to corresponding main integrators (horizontal, vertical and end-offield) by increasing the voltages in the latter until equal to those in the former. The horizontal and vertical main integrators supply the two deflection voltages to the CRT, and now contain the initial positioning voltages. Current can now be fed to these two integrators direct to cause scanning of the beam from the initial . position. The end-of-field main integrator is used during field reading to stop the scan at the end of the field, its voltage being compared with that produced by the horizontal main integrator for this purpose. Error features.-To compensate for changes in dimensions (due, e.g. to humidity) and bad positioning of the document including skew, each document (Fig. 6, not shown) has a number of pairs of reference lines printed on it, one horizontal and one vertical line in each pair, and the computer may order one or more error measurements before any field reading operations. In a typical such measurement, the CRT beam is positioned at a reference location near such a pair of lines and then moved in turn horizontally until it meets the vertical line and vertically until it meets the horizontal line, the beam being backspaced slightly between these operations. The meetings referred to are detected by a photomultiplier tube viewing the document. During each of these two movements the voltage of a corresponding integrator, preset to a particular value, is progressively altered until the meeting with the reference line. The residual voltage will be zero in the case of no error and, if non-zero, biases one or more of the " format " integrators to correct for the errors. If the meeting with a reference line is too long delayed, indicating that it has probably been missed altogether, scanning is terminated automatically. The pairs of reference lines used for error measurements depend on document length as specified by the computer. Further features.-The system may also perform a "scan and count" " operation (not described in detail) for diagnostic purposes and for developing a calibration chart for the CRT by scanning test documents. It is mentioned that the first character of a control word may additionally specify when to interrupt a scan or identify the type of characters (on the document) to be recognized. Movement of the beam out of the viewing window is detected by continuous comparison of the deflection voltages with reference voltages. An ageing mode of operation of the system is possible, wherein the beam scans back and forth across the CRT screen, this mode being temporarily entered if an operation terminates before the voltages for the next operation have completely transferred from the " format " integrators to the main integrators.
-
公开(公告)号:DE1524431A1
公开(公告)日:1970-08-20
申请号:DE1524431
申请日:1966-11-16
Applicant: IBM
Inventor: LEE MALABY DAVEY
Abstract: 1,156,229. Centring character scanner. INTERNATIONAL BUSINESS MACHINES CORP. 28 Oct., 1966 [24 Nov., 1965], No. 48333/66. Heading G4R. Scanning apparatus comprises means for scanning a character and obtaining binary data signals each representing only the presence or absence of an element of the character in a respective segment, defined independent of the character, of the scan field, means for modifying the data signals in accordance with a predetermined criterion independent of the character and comparison means for comparing the modified data signals and providing an output signal which tends to centre the scan field with respect to the character. A flying-spot scanner can scan a document in a vertical raster with either of two horizontal spacings, feeding the scanned bits to a shift register. Initially the wide spacing is used travelling rightwards (mode 0) to locate the first character of a line of characters. Detection of two consecutive black bits in the first two stages of the shift register at any time during a vertical scan indicates the character has been reached and causes the narrow spacing to be used, continuing rightwards (mode 1). During this mode, used for centring the scans vertically on the line, the contents of the first four stages of the shift register are sampled at appropriate times during each scan to set latches on meeting the so-called top, centre and bottom conditions and to increment/decrement a threshold counter, the latches and counter being reset after each scan. The top condition is met if black bits are detected in bit positions 6 and 7, or 7 and 8, in the scan. The centre and bottom conditions relate similarly to positions 11-13 and 23-25, respectively. The threshold counter is decremented by 1, 1 and 2 in response to black bits in scan positions 2, 3 and 4, respectively, and incremented by 2, 1 and 1 in response to black bits in scan positions 27, 28 and 29, respectively. At the end of the scan, the vertical position for the next scan is lowered by one bit position if either the count is equal to or greater than + 1 or the top and centre conditions have not both been satisfied. The vertical position is raised by one bit position if the top and centre conditions have been satisfied and either (a) the count is less than or equal to -1 or (b) the magnitude of the count is less than 1 and the bottom condition has not been satisfied. If the magnitude of the count is less than 1 and the top, centre and bottom conditions have all been satisfied, mode 1 ends and mode 2 follows. In mode 2, the wide raster spacing is used, continuing rightwards. The threshold counter is fed as before but is only reset every 7 scans. At the end of every 7 scans, the vertical position for the next scan is raised one bit position if the count is less than or equal to - 2 and lowered one bit position if the count is greater than or equal to + 2. When a pre-printed right margin stripe on the document is reached, mode 2 ends and mode 3 commences. During mode 3 the scanning raster returns leftwards using the narrow spacing, and recognition circuitry responds to the shift register contents to identify every character reached. Vertical centring on the line of characters occurs as in mode 2 except that the threshold counter is fed during the first 15 scans of each character only and the count is used to control centring, and reset to zero, only between characters.
-
公开(公告)号:DE1948387A1
公开(公告)日:1970-07-02
申请号:DE1948387
申请日:1969-09-25
Applicant: IBM
Inventor: LEE MALABY DAVEY , ANDREW WESLEY MICHAEL
IPC: G06F7/00 , G06F7/02 , G06F7/575 , H03K19/177 , A01D43/10
Abstract: 1,238,273. Data processing. INTERNATIONAL BUSINESS MACHINES CORP. 1 Oct., 1969 [14 Oct., 1968], No. 48213/69. Heading G4A. Data processing apparatus comprises an M x N matrix array of cells 10-MN, an output D of each cell (except the last) in a row being applied as an input B to the next cell in its row, an output E of each cell (except the last) in a column being applied as an input C to the next cell in its column; two storage stages U, V associated with each column for determining the operation to be performed by the column; two storage stages X, Y associated with each cell for determining the operation to be performed by the cell in accordance with the operation determined by the U and V stages; means for applying inputs B 1 -B M to the first column, inputs A 1 -A M to respective rows and inputs C 0 -C N to the first row; the array producing outputs D 1 -D M from the last column and outputs E 0 -E N from the last row. The stages U, V, X, Y are flip-flops and are connected as a shift register for each column (or alternatively the U, V, X, Y stages could be an addressable memory, or could be controlled by respective light-sensitive diodes or transistors illuminated via a changeable mask). Fig. 6 shows how the outputs D, E of a cell depend on its inputs A, B, C for the various values of U, V, X, Y, to permit data routing down and between columns, and logical and arithmetic operations. The arithmetic operations are addition of 0 or 1 to the column (according to the value of V). Multiorder binary numbers can be obtained at E 0 -E N from addition, subsequent to propagation of sum bits down columns and carry bits along rows. Threshold functions can be achieved, as can computation of binary or ternary dot products with production of mismatch counts. A plurality of arrays as above can be connected together by connecting D to B and E to C, or E to A. Corresponding inputs A and B (in a given array) can be commoned together, as can corresponding inputs V and C.
-
公开(公告)号:DE1524412A1
公开(公告)日:1971-04-01
申请号:DE1524412
申请日:1966-08-02
Applicant: IBM
Inventor: ANDREW GARRY GERALD , JAMES KOSTUCH DONALD , ROCHESTER MINN , WESLEY LAWLESS FREDERICK , LEE MALABY DAVEY
Abstract: 1,109,349. Line identification. INTERNATIONAL BUSINESS MACHINES CORPORATION. 25 July. 1966 [9 Aug., 1965], No. 3309/66. Heading G4R. Line identification apparatus detects coded line representations fixed with respect to a document which bears lines to be identified. Fig. 4 shows a document which bears lines of characters (not shown) to be scanned and read by a cathode ray tube (and photomultiplier),. Sets of coded indicia in a line identification track are provided to allow lines to be identified for subsequent marking when they contain unreadable characters, and to facilitate indexing of the document after one group of lines has been scanned to bring the next group into position. Each set of coded indicia is a set of horizontal bars 19 present or absent to represent 1 and 0 respectively. Thicker so-called bracket bars 17 separate successive sets of the coded indicia. Other bars are provided as shown. The document is advanced until a photodetector 33 has detected both bars 30 and 29 after which it is stopped. The beam of the cathode ray tube is then at 35. The beam is moved rightwards until it encounters bar 27 when it backs off and moves downwards to read coded indicia 26 (bars present and absent) identifying the document. When a predetermined distance below a bracket bar 28, the beam moves rightwards in a line searching raster to find a line of characters. On detecting a line, the raster is normalized and centred on it (no details) as it moves rightwards. When it reaches the right-hand end of the line it returns leftwards to read the characters (no details). If all the characters are read successfully, the beam moves downwards and rasters along the next line. However, if a character could not be read the beam crosses horizontally into the line identification track then moves upwards. During the upward movement until the first bracket bar is detected, a first counter (354, Fig 2e, not shown) counts clock pulses to give a fine component of the line position, this fine component then being transferred to latches (363- 368, Fig. 2e, not shown). The coded indicia detected until the next bracket bar is reached give a coarse component of the line position and are gated into latches (377-382, Fig. 2e, not shown) under control of the decoded (410) output of the first counter (354) now being fed with clock pulses for this purpose. The fine and coarse components are passed to a central processor store. The beam returns downwards along the same path (despite Fig. 4) and after it passes the lower bracket bar, the first counter (354) counts clock pulses until the count equals the latched fine component (363-368) as determined by AND gates (395-401, Fig. 2e, not shown) whereupon the beam moves rightwards to find another line of characters. For indexing the document, the coarse and fine components of the position of the last line to be scanned are obtained and latched as before, the coarse component being passed to further latches (389-394). The document is moved and bars 16, 17 detected at 39 are counted by a second counter (427, Fig. 2f, not shown) preset with the number of bars between the detector 39 and the required stop position. When the count equals the latched coarse component (389-394) as determined by AND gates (428-433, Fig. 2#, not shown), clock pulses are gated to the first counter (354, Fig. 2e, not shown) the count being compared (395-401) with the latched fine component (363-368) of the last line to be scanned prior to indexing. On equality, the document is stopped. The last line to be scanned prior to indexing is now identified by reading successive sets of the coded indicia during downward movement of the beam, latching them (377-382, Fig. 2e, not shown) and comparing them (402-107, Fig. 2f, not shown) with the coarse component (389-394) of said last line. On equality, beam movement continues with counting of clock pulses in the first counter (354, Fig. 2e, not shown) until the count equals (395-401) the fine component (363-368) of said last line. The beam then moves downwards and to the right to scan the next line of characters. Lines can be marked by printing in the line mark track 36 at another station while a further document is being read. As the document moves through the second station, a detector 67 detects the bars 16, 17 which are counted (475, Fig. 2b, not shown) the count being compared (476-482) with the coarse component of the position of the first line to be marked (from the central processor store). On equality, clock pulses are gated to a further counter (485, Fig. 2b, not shown) until the count equals (486-492) the fine component of the line position (from the central processor store) when the marker is actuated. Further lines are marked similarly. Modifications.-In a second embodiment, the document (Fig. 7, not shown) is similar except that thick and thin (rather than present and absent) bars are used to represent 1 and 0, and a bar (56) to stop the document initially is in the line identification track. An 8-bit shift register (515, Figs. 5c, 5g, not shown) is provided for receiving a 5-bit set of document or line identifying indicia (read during downward movement of the beam) and two 1-bits produced by two bracket bars bracketing the set. An AND gate (548, Fig. 5g, not shown) verifies that a valid code has been shifted in, viz. that the first and seventh stages hold 1 (the bracket bars) and the eighth holds 0, and a trigger (550, Fig. 5g, not shown) verifies that an odd number of bits have been shifted in. A fine component is obtained by counting (595, Fig. 5d, not shown) as before. Comparisons of fine component counts are done by entering the complement of the first into the last-mentioned counter (595) and incrementing to capacity. The line identification track could be on a belt or plate referenced with respect to the document, rather than on the document itself.
-
公开(公告)号:DE1499394A1
公开(公告)日:1970-01-02
申请号:DE1499394
申请日:1965-12-09
Applicant: IBM
Inventor: LEE MALABY DAVEY
Abstract: 1,060,920. Character recognition. INTERNATIONAL BUSINESS MACHINES CORPORATION. Nov. 22, 1965 [Dec.18, 1964], No. 49453/65. Heading G4R. Character scanning apparatus comprises a scanning device, means for measuring the centreline height of a character to be scanned, and means for utilizing said measurement for controlling the vertical limits of the scan. In character recognition apparatus, bit signals from vertical scans of a character by a cathoderay tube are passed to a shift register (100). The following conditions are looked for in turn during each scan, detection of a condition setting a corresponding latch (154, 158, 160, 162, 164) and resetting the latch of the preceding condition: (a) the pattern X00011 in the shift register (X being a " don't care " bit) indicating the crossing of the lower edge of the lowest line of the character; (b) the pattern XXXX11 indicating the crossing of the upper edge of the lowest line; (c) a count of 14 bit segments in a " skip " counter (225) enabled to count clock pulses while conditions (a) and (b) are satisfied; (d) the pattern X00011 indicating the crossing of the lower edge of the highest line of the character; (e) the pattern XXXX11 indicating the crossing of the upper edge of the highest line. If conditions (b) and (e) are not satisfied in five bit times after satisfying conditions (a) and (d) respectively, the (a) and (d) latches (154, 162) are reset by an ABORT signal following on recognition of pattern 11XXX1 or 1X1XX1 and conditions (a) and (d) have to be satisfied again if this scan is to be used for the purpose described below. In addition the (a) condition latch (154) can only be set during the first half of the scan. While the condition (a) and (d) latches (154, 162) are set, a " twice centreline " counter (250) counts one each bit time and while the condition (b) and (c) latches (158, 160) are set, it counts two each bit time. Provided the condition (e) latch (164) is set shortly following the end of a scan, six " centreline " counters (402, 410, 418, 426, 434, 442) are then enabled to respond to the " twice centreline " counter (250) as follows. Six lines (N15 to N20) are provided corresponding to " twice centreline " counts of 15 to 20 respectively. If the " twice centreline " count is even, the line corresponding to half the count is marked, whereas if it is odd the two lines corresponding to adjacent numbers which sum to give the count are marked. Each line (N15 to N20) corresponds to a respective one of the " centreline " counters and when marked will increment by one all the " centreline " counters not at zero, provided the counter corresponding to the line is at zero, whereas if this counter is not at zero, it (alone) is decremented by one. After a number of scans covering the whole character as recognized by a scan counter (477), the identity of the " centreline " counter at zero indicates the maximum character centreline height and a corresponding number is entered into a 4-stage binary " intermediate control " counter (550). The latter and a similar " control " counter (570) which was preset to ten, are then incremented until the former (550) reaches 0000, provided the circuitry had attempted to set the lowest order trigger (551) of the " intermediate control " counter (550) to either 1 or 0. The " control " counter reading is converted to an analogue voltage in weighted resistors and used to augment a standard CRT deflection voltage subsequently used for recognition scanning.
-
-
-
-