×

Neural network for character recognition of rotated characters

  • US 5,319,722 A
  • Filed: 10/01/1992
  • Issued: 06/07/1994
  • Est. Priority Date: 10/01/1992
  • Status: Expired due to Term
First Claim
Patent Images

1. A method for transforming a circularly arranged character to a linearly arranged character, comprising the steps of:

  • generating a plurality of address locations for grey scale sampling pixel values making up said circularly arranged character, said address locations being arranged as a polar array of points formed by intersections of arcs and radial lines, said radial lines extending from a center point of the circle about which said circularly arranged character is arranged, and said arcs are constructed about said center point;

    generating an address array storing addresses of said sampling points;

    sampling said grey scale pixel values of said circularly arranged character, the number of sampling points being greater than the number of pixels making up the circularly arranged character, and wherein said radial lines are spaced by an integer divisor of the vertical height of a pixel and said arcs are spaced by an integer divisor of one horizontal width of a pixel;

    storing said sampled grey scale pixel values in a target array;

    constructing said linearly arranged character from the sampled values;

    comparing said sampled pixel values with a reference value and assigning binary values to said pixel values on the basis of said comparison;

    after said comparing step, finding a center of said linearly arranged character and finding an edge of said linearly arranged character;

    wherein said step of generating address locations includes computing an address location as x and y coordinates and rounding said coordinate off to the nearest pixel address in the circularly arranged image; and

    applying said linearly arranged character to a neutral network.

View all claims
  • 2 Assignments
Timeline View
Assignment View
    ×
    ×