Library image optimization

• US 4,887,304 A
• Filed: 03/16/1989
• Issued: 12/12/1989
• Est. Priority Date: 09/30/1987
• Status: Expired due to Fees

• Current Assignee: Raytheon Company (Rtx Corporation)
• Original Assignee: Raytheon Company (Rtx Corporation)
• Inventors: Terzian, John
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): COUSO, YON JUNG
• Application Number:
  US07/326,436
• Time in Patent Office: 271 Days
• Field of Search: 382/30, 382/34, 382/36, 382/33, 382/15
• US Class Current: 382/209
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries

Abstract

A process is disclosed for forming reference images useful in object identification processes where an image of an unknown object is compared to each image in a library of reference images. The disclosed process forms the reference image by combining a plurality of images of objects of the same class.

Claims

1. In a process for identifying an unknown object by comparing a received image of the unknown object with each one of a plurality of reference images in a library of reference images, each reference image in the library being representative of an object of known classification, a method for forming each reference image in the library comprising the steps of:

(a) forming an initial library containing a plurality of reference images, each such reference image being the image of an object of an identifiable class of objects, and each such image being represented by a plurality of vertical vectors;

(b) generating a set of additional reference images, each of the reference images in the set being a different image of objects of the same class, and each such image being represented by a plurality of vertical vectors, the vectors in each such image being ordered such that each vector in each image corresponds to one vector in each of the other additional reference images; and

(c) computing a plurality of scores for each image in a set of additional reference images, each score representing the degree of similarity between the additional reference image and a different one of the reference images in the initial library and all images in the set of additional reference images being formed from objects in one class which the process for identifying an object can identify;

(d) forming a correlation margin for each image in the set of additional reference images, such correlation margin indicative of the degree to which the additional reference image differs from the reference in the initial library;

(e) forming an accumulated composite image by adding corresponding vertical vectors in all of the additional reference images with correlation margins in a range of values; and

(f) forming a new library image by combining the accumulated composite image with the reference image in the initial library of the object of the same class as the objects used to form the additional reference images.

2. The process for forming a reference image as in claim 1 additionally comprising the steps of:

(a) replacing the reference image in the initial library of the object of the same class as the objects used to form the additional reference images with the new library image; and

(b) repeating steps (a) through (f) of claim 1.

3. A process of forming a library of reference images wherein the process of claim 1 is repeated a plurality of times, each repetition done with a set of additional reference images of a different class of objects.

4. The process for forming a reference image as in claim 1 wherein the scores rank from lowest to highest in order of degree of similarity between the additional reference image and the reference image in the initial library, and (a) when the lowest score is computed for the reference image in the initial library of the object of the same class as the object used to form the additional reference image, the correlation margin computed for each additional reference image equals the difference between the second lowest score for the additional reference image and the lowest score for that additional reference image;

and (b) when the lowest score is not computed for the reference image in the initial library of the object of the same class as the object used to form the additional reference image, the correlation margin computed for each additional reference image equals the difference between the lowest score for that additional reference image and the score for that additional reference image computed for the reference image in the initial library of the object of the same class as the object used to form the additional reference image.

5. The process for forming a reference image as in claim 12 wherein the accumulated composite image and a reference image in the initial library are combined to form a new library image by combining each vector in the accumulated composite image with the corresponding vector in the reference image in the initial library to form the corresponding vector in the new library image according to the formula