中科微点-全球专利检索

  1. Login
  2. Sign Up

Search Results

15 records (took 0.014 seconds)

    Data conversion techniques for producing autocorrelation functions
    1.
    发明授权
    Data conversion techniques for producing autocorrelation functions 失效
    用于产生自相关函数的数据转换技术

    公开(公告)号：US3413602A

    公开(公告)日：1968-11-26

    申请号：US40326264

    申请日：1964-10-12

    Applicant: IBM

    Inventor： HORWITZ LAWRENCE P ,  SHELTON JR GLENMORE L

    IPC: G06F17/15 ,  G06K9/52 ,  G06K9/74 ,  G06K9/80

    CPC classification number: G06K9/745 ,  G06F17/15 ,  G06K9/52 ,  G06K9/522 ,  G06K9/64 ,  G06K9/80

    Abstract: 986,276. Automatic character reading. INTERNATIONAL BUSINESS MACHINES CORPORATION. Feb. 26,1962 [March 3, 1961], No. 7423/62. Heading G4R. In character recognition apparatus scanning means provide a series of signals representing the character, and further means provide from this series, signals representing the autocorrelation function of the series, the autocorrelation signals being further modified and used to recognise the character. The character 1, Fig. 1, is scanned in a raster by a cathode ray tube 5, signals from photo-cell 7 being applied to an autocorrelation function generation 9. This consists of a 22-stage shift register the photo-cell signals being gated into the register by 45 shift pulses, one for each of the 5 x 9 cells of the character area. For each stage of the register there is a gate connected to the input, so that as the character signals step through the register each gate produces a number of pulses representing the number of overlaps the original character pattern 1 would make with itself in various relative positions, a gate corresponding with each displacement position. The pulses are passed to a second-difference operator circuit 11 in which the table of numbers Fig. 11 representing the autocorrelation function is multiplied by a second-difference operator to increase the black/white discrimination. Each gate lead is connected to an amplifier giving positive and negative outputs. These outputs pass to integrators which sum the sequence of pulses. The second-difference operator is applied by passing the positive amplifier outputs to the corresponding integrators through resistors valued "1" and the negative amplifier outputs to integrators corresponding to the four neighbouring cells (above,below and on each side) through resistors weighted 4. By this means the positive signal in the central position is four times the negative signals in the other four positions. The integrators sum the succession of signals and their final outputs give the seconddifference junction Fig. 11. Each output is applied to a threshold circuit 13 giving a three-state signal depending upon whether the signal was greater, equal to or less than zero to produce the final table as shown in Fig. 12a. Amplifiers produce positive and negative versions of the input which are connected through equal resistances to ten output leads. Each lead is connected to the positive outputs of those amplifiers which correspond to positions having "1" in the table,and the negative outputs of those amplifiers corresponding to "-1". No connection is made to the amplifiers corresponding to positions having zeros in the table. Each character lead receives a signal which depends upon the match between its non-linearly weighted second-difference function and the corresponding signals produced from the scanned character. Because the characters are of different areas,the maximum signals, i.e. the signals generated in the corresponding leads by the ten characters, are different. They must therefore be normalised before being applied to the highest signal detector. This is done by connecting each character lead to earth through a suitably weighted resistor. A transistor circuit 17 finds the lead having the highest signal, the transistor connected to the highest lead conducts and prevents conduction of those connected to any other leads. If two transistors should conduct a reject circuit responds to reject the character. One of ten lamps may light to indicate the character identified. In practice many more cells than 45 are necessary and the circuitry is correspondingly multiplied. The circuit is described in detail with reference to Fig. 22 (not shown).

    Pattern recognition preprocessing techniques
    8.
    发明授权
    Pattern recognition preprocessing techniques 失效
    模式识别预处理技术

    公开(公告)号：US3339179A

    公开(公告)日：1967-08-29

    申请号：US53850466

    申请日：1966-02-23

    Applicant: IBM

    Inventor： SHELTON JR GLENMORE L ,  HORWITZ LAWRENCE P

    IPC: G06K9/44 ,  G06K9/56

    CPC classification number: G06K9/56 ,  G06K9/44 ,  G06K9/46 ,  G06K2209/01

    Abstract: 981, 500. Automatic character reading. INTERNATIONAL BUSINESS MACHINES CORPORATION. May 20, 1963 [May 21, 1962], No. 19944/63. Heading G4R. In a character reading apparatus a signal is derived from the signals obtained by scanning the character by accepting all portions having a magnitude greater than a predetermined value and also those portions having a magnitude below this value when they are needed to ensure that the output signal defines a continuous line pattern. The signal derived by scanning a character by means of a flying-spot scanner may be shown as in Fig. 2A, the number representing the magnitudes of the original produced at each sample point. To eliminate all signals below 2 would discard necessary data and leave an incomplete pattern. The invention retains data where it is necessary to provide a continuous pattern line and discards superfluous data on each side of the line. The pattern obtained is shown in Fig. 2B. A decision as to whether any particular point is necessary is made in logical circuits controlled by the presence or absence of signals in the surrounding points. A matrix of nine points is considered as shown in Fig. 4A, the unknown point X being in the middle. Point X, if absent, is taken as being present if it is required to connect other points, say A and C, which would otherwise not be connected, because the signal in the B position is absent as the result of a previous operation of this stage. Some of the logical conditions which necessitate a signal to be present at X are shown in Fig. 4A. The circuit is shown in Figs. 5A, 5B, 5C. The character is scanned by a flying spot from C.R.T. 6 and photo-cell 7 provides corresponding signals for application to a gated amplifier 15 in which they are sampled by pulses synchronized with the scan of the C.R.T. 5. The samples, corresponding to successive points on a co-ordinate array as shown in Figs. 2A, are applied to a threshold device T1 with a low level e.g. about 0. 5 where the signal varies from 0 to 3. A series of "1's" are accordingly entered, in vertical scans, into a shift register 19 and outputs are taken from positions so spaced that they correspond to the points A, B, C, D, E, F, G and H in Fig. 4A. The shift register must contain simultaneously the data relating to three vertical scans, with say 20 sample points in each. The outputs are applied to connectivity function circuit 37 along with outputs from the next shift register 21 which stores decisions made on previous points in the circuit 37. The output of circuit 37 which indicates whether X is to be taken as "1" because it is necessary for line continuity or "0" because it is not, is applied to an OR gate 39. The other two inputs indicate the "blackness" range of the point X which has been suitably delayed at 41 and applied to threshold circuits 43, 45 "T 1 -low" and "T 1 -high". These may be set at values say ¢ and 1¢. The output of circuit 43 is inverted at 44 so that, between them, these two circuits provide a signal for OR gate 39 if the magnitude of the signal for point X is below or above the magnitude range for which the first stage is set. In these two cases of if the point X is required for line continuity the gated amplifier 49 is enabled to pass the corresponding magnitude signal to the next stage. The input to the next shift register is through a higher-level threshold device 17 and other circuitry as the same as in the first stage. Signals are passed on to the Nth stage and so on, lower magnitude signals being successively eliminated unless they are required for connection purposes so that the final signals represent the array of dots as in Fig. 2B. The connectivity function circuits consist of combinations of inverters and gates. In addition there is an adding and threshold circuit which provides an output for the X position if signals are present at more than a certain number (e.g. 6) of the surrounding positions.

Patent Agency Ranking