953,442. Automatic character reading. INTERNATIONAL BUSINESS MACHINES CORPORATION. April 5, 1960 [April 8, 1959], No. 11959/60. Heading G4R. In a character reading apparatus the character 12 Fig. 1 is sensed by a row of adjacent reading heads 16 covering a broad band e.g. twice the depth of the character and the reading heads are connected to a channel reduction circuit 20 consisting of a number of Or gates equal to the number of heads required to cover a character, no two heads sensing a character being connected to the same gate. The character is printed in magnetic ink and is pre-magnetised by an A. C. head at, say, 30 kc. and signals of this frequency are induced in the read-heads 16. If the character is covered by ten heads, there are twenty altogether to provide a double width scan and the heads are connected in pairs, the first and eleventh, the second and twelfth and so on to provide a complete character signal on only ten heads. A typical signal on one of these heads is shown in Fig. 2a at I. After amplification in a transistor amplifier adapted to give a high signal to noise amplification (22, Fig. 1) the signal is full-wave rectified (II, Fig. 2a) and clipped in a transistor threshold device 24, Fig. 1 to give the squared wave shown at III. This wave is smoothed in block 26 by being delayed (IV, Fig. 2a) and the two resulting waves mixed in an Or gate. Long duration D. C. pulses are thereby produced representing those parts of the scan where the character is detected i.e. the black areas. These pulses are used to cause a transistor to conduct which discharges a condenser at a steady rate and when a predetermined threshold level is reached a trigger is set which recharges the condenser and blocks the discharging transistor. This is in effect an integration of the pulses (28, Fig, 1) and a "digital" trigger 30 is set when a sufficiently long black signal is obtained. To avoid response to spurious signals it is reset if it is not followed by the setting of a second digital trigger within a very short time afterwards. The integration process is restarted at six evenly spaced points in the scan and the digital trigger is tested in each period to determine whether it has been set. If it has a buffer trigger 34 is set. The six timing pulses are derived from a clock circuit 32 the operation of which is initiated when at least two out of the ten digital triggers are set. This is sufficient evidence that the leading edge of a character has been detected. The presence of a document 10 is detected by a photocell, switch or the like to set various triggers on the timing circuit to their proper conditions in readiness for a recognition operation. The output of the timing circuit is in effect gated with the set output of the digital trigger to set the buffer trigger each time a digital trigger has been set in a scan period or character increment XII, Fig. 2a. The sensed signals are entered into a matrix of triggers having ten rows and seven columns one for each character increment. Since the timing circuit 32 does not start until the first black signals arrive, these signals cannot be entered via the buffer triggers and are therefore entered directly into the first column of triggers from the digital triggers. Successive increment signals are entered into the following six columns under the control of timing pulses appearing successively on successive column lines. According to the position of the character in relation to the twenty reading heads the upper part of the character may be represented at the upper part or the lower part of the matrix. To enable a recognition to be made, the trigger settings are displaced vertically, the top row being connected to the bottom row and ten shift pulses being applied to roll the stored pattern right through the matrix back to its original position. The character is centered in the matrix when the top or bottom row of triggers are all in the "white" condition, and the recognition circuit is enabled when this condition is detected as described below. The position in the train of ten pulses where the recognition of a character occurs is stored and for the next following character the recognition circuit is enabled for five pulses centered about that position. A special symbol may be printed in front of each row or block of characters which is designed to be easily recognised so that the recognition position may be established to facilitate the timing of subsequent recognition samplings. If more than one recognition output is obtained for each character, an error circuit gives a warning. If no recognition is obtained a similar warning is given. Recognition. A charater is recognised by looking for stored black and white signals at particular places in the matrix when the stored image comes into a centered position. The signals expected for a character "2" are shown in Fig. 11. The characters are stylised as shown in Fig. 1 and a "2" would ideally give a row of four black positions in row 2, black in positions 39, 49, black in positions 5D-59, black in positions 6D and 7D, black in positions 8D-89 and all the rest white. However it is sufficient to ignore most of the seventy positions and look for white or black signals only in the positions marked. Where two positions are connected together with a line a signal in either position will do. In Fig. 10 set and unset outputs of appropriate triggers of the matrix are gated together, the alternative positions in Or gates 271-278 and then the single positions and the Or gate outputs in an And gate 279. Or gates 271-274 detect the presence of white in the top or the bottom rows and indicate that the character is properly located in the matrix. Recognition trigger 281 is enabled by the five recognition sampling pulses referred to above applied on terminal 283.