×

Machine learning procedures for generating image domain feature detector structuring elements

  • US 4,821,333 A
  • Filed: 08/22/1986
  • Issued: 04/11/1989
  • Est. Priority Date: 08/22/1986
  • Status: Expired due to Fees
First Claim
Patent Images

1. A method for generating a set of structuring elements for use in an image processing program in a computer, the image processing program being operative to apply feature detectors to an input image comprising an input digital matrix of points having digital values to generate a corresponding transformed digital matrix of points having digital values, each point of said transformed digital matrix having a digital value which is a function of the digital value of a corresponding point in said input digital matrix and/or the digital values of points immediately neighboring said corresponding point in said input digital matrix, the transformed digital matrix containing either no points having nonzero values or at least one point having a nonzero value, each featuare detector consisting of a program form including a predetermined series of morphological operations applied to the input digital matrix and a spatial constant component comprising a set structuring elements, the method comprising the steps of:

  • (a) selecting a plurality of sets of structuring elements;

    (b) representing each set of structuring elements in a single binary code word chromosome in accordance with a predetermined encoding technique, thereby producing a plurality of said chromosomes;

    (c) applying a plurality of featuare detectors equal in number to the number of said plurality of chromosomes to a first set of images known to belong to a predetermined class to and a second set of images known not to belong to said predetermined class, each feature detector consisting of said program form and a unique one of said plurality of chromosomes, each feature detector thereby producing a transformed digital matrix containing either no points having nonzero values or at least one point having a nonzero value;

    (d) associating a score with each feature detector based upon the number of input images from said first set of input images in which said feature detector produces a transformed digital matrix containing at least one point having a nonzero value and the number of input images from said second set of input images in which said feature detector produces a transformed digital matrix containing no point having a nonzero value;

    (e) determining whether the score for any of said plurality of feature detectors meets a predetermined criterion;

    (f) if the score of at least one of said plurality of feature detectors meets said predetermined criterion, identifying the set of structuring elements represented by the single binary code word chromosome for one such feature detector having a score meeting said predetermined criterion said identified set of structuring elements being the set of structuring elements generated by the method;

    (g) if the score of none of said plurality of feature detectors meets said predetermined criterion(1) producing a succeeding plurality of chromosomes representing a succeeding plurality of sets of structural elements by randomly selecting a subset of said plurality of chromosomes where the probability of selection of any chromosome of said plurality of chromosomes is proportional to the score associated with the feature detector employing that chromosome, and augmenting said subset of chromosomes with chromosomes corresponding to modifications and/or combinations of chromosomes of said subset of chromosomes,(2) repeating said step of applying a plurality of feature detectors to said first set of images known to belong to said predetermined class and to said second set of images known not to belong to said predetermined class employing said succeeding plurality of chromosomes,(3) repeating said step of associating a score with each feature detector, and(4) repeating said step of determining whetherthe score for any of said plurality of featuredetectors meets said predetermined criterion, until the score of at least one of said plurality of feature detectors meets said predetermined criterion.

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