PREPROCESSING METHOD AND APPARATUS FOR PATTERN RECOGNITION

• US 3,636,513 A
• Filed: 10/17/1969
• Issued: 01/18/1972
• Est. Priority Date: 10/17/1969
• Status: Expired due to Term

• Current Assignee: Glenn E. Tisdale
• Original Assignee: Glenn E. Tisdale
• Inventors: Tisdale, Glenn E.
• Primary Examiner(s): Robinson, Thomas A.
• Application Number:
  US04/867,250
• Time in Patent Office: 823 Days
• Field of Search: 340/146.3
• US Class Current: 382/204
• CPC Class Codes: G06V 10/757 Matching configurations of points or features
• CPC Class Codes: G06V 30/248 involving plural approaches, e.g. verification by template match; Resolving confusion among similar patterns, e.g. "O" versus "Q" G06V30/242 takes precedence

Abstract

Features are extracted from a two-dimensional image for subsequent classification of patterns within the image according to correspondence between the extracted features and reference features in a set extracted previously from known patterns. In extracting the features, measurements are first taken of observed characteristics of the image about two or more predefined points in the image, these measurements being chosen to be invariant regardless of orientation, scale, and position of the pattern in the image. The measurements, along with data regarding relative positions of the selected points, constitute the features from which eventual pattern recognition may be achieved.

Claims

1. The method of preprocessing information contained in an image to permit classification of an unknown pattern that may be present within said image relative to a set of reference pattern classes, which comprises:

accepting certain points within the image of substantial information-bearing character as image points for the extraction from the image of features of any patterns contained within the image. taking measurements with respect to acceptable image points, said measurements being chosen to be invariant regardless of orientation, scale, and position of any pattern associated with each such image point, and extracting as features of said pattern the values of said invariant measurements together with sufficient data to establish the geometrical relationship between the accepted image points, for comparison with similarly determined features in each of the reference pattern classes.

2. The method of claim 1, wherein each said image point is accepted as representative of a predominant characteristic within said pattern.

3. The method of claim 1 wherein at least some of said measurements are taken to establish the orientation of lines emanating from said image points relative to the line of direction between a pair of said image points.

4. The method of claim 1 wherein at least some of said measurements are taken to ascertain the color or intensity of color at said image points.

5. The method of claim 1 wherein at least some of said measurements are taken to establish the maximum gradient of gray scale value relative to some of said image points.

6. The method of claim 1 wherein measurements for each feature are taken about only two image points, and the number of features to be extracted is n(n- 1)/2, where n is the totality of points available for feature formation.

7. The method of claim 2 wherein said measurements are taken to establish the orientation of lines emanating from at least some of said image points relative to the line of direction between a pair of said image points, said lines being selected to correspond with actual lines or gradient contours in said pattern.

8. The method of claim 1 wherein the information is contained in two or more images derived from sensing devices of different spectral response.

9. The method of claim 1 wherein two or more images are under consideration which pertain to the same field of view but which have been derived from different vantage points relative to that field of view.

10. A process for extracting features contained within a two-dimensional image for subsequent recognition of an unknown pattern that may be present within the image as one of A set of known patterns, said process including performing measurements of observed phenomena about two or more selected points in the image, which measurements are independent of scale, orientation, and position of any pattern with which they may be associated, detecting the relative positions of said selected points in the image, and utilizing the information obtained from the invariant measurements, and data indicative of the relative positions of the points about which said measurements are taken, as features of unknown patterns that may be present in the image, from which pattern recognition may be achieved.

11. The process according to claim 10 wherein said measurements are of color.

12. The process according to claim 10 wherein said measurements are of intensity of gray scale value.

13. The process according to claim 10 wherein said measurements are of maximum gradient of gray scale intensity relative to said points.

14. The process according to claim 10 wherein said measurements are of orientations of lines emanating from some of said points relative to a line joining a pair of said points.

15. The process according to claim 14 wherein said lines emanating from selected points coincide with lines in the image.

16. In a pattern recognition process, the steps of:

scanning a two-dimensional image to extract features of unknown patterns that may be present within the image suitable for classifying each such pattern, detecting in the scanned image selected measurable characteristics that are invariant regardless of scale, orientation, or position of any unknown patterns with which those characteristics may be associated within the image. measuring at least one of said characteristics at a plurality of points in said image, and extracting the measured data along with data indicative of geometrical relationships of the points at which the measurements are taken to determine the relative positions of the points, as features for classifying the unknown patterns, if present.

17. In a pattern recognition process, the steps of:

performing measurements about points of high information content in an image, which measurements are invariant with respect to orientation, scale, and position within the image of a pattern including said points, determining the distance between pairs of said points, and the orientation of a straight line containing a pair of the points relative to a reference axis, and supplying information representative of the invariant measured values and of the distance between pairs of points and the orientation of said line by which to detect the known pattern class in which lies an unknown pattern associated with at least some of said measured values and points.

18. Apparatus for extracting information from a field of view preliminarily to classification of any unknown pattern that may be present within said field of view, comprising:

means for detecting information characteristic of and peculiar to said field of view, means responsive to the detected information for deriving therefrom a smaller amount of information of substantial content representative of prominent characteristics within said field of view, in contrast to indistinct background of low information content, means responsive to said information of substantial content for obtaining measurements with respect to points of prominence chosen therefrom, said measurements being invariant regardless of orientation, scale, and position of an unknown pattern that may be present within said field of view and associated with said points, and means responsive to said measurements for determining the relative positions of pairs of said points for subsequent use with said invariant measurements to classify the associated unknown patterns, if present.

19. Apparatus as recited in claim 18 wherein said detecting means comprises at least two distinct detecting means having different spectral responses.

20. Apparatus as Recited in claim 18 wherein said detecting means comprises at least two distinct detecting means exposed to the same field of view but having different vantage points relative to the field of view.