Pattern recognition using multiple templates

• US 6,766,058 B1
• Filed: 08/04/1999
• Issued: 07/20/2004
• Est. Priority Date: 08/04/1999
• Status: Expired due to Fees

• Current Assignee: Electro Scientific Industries Incorporated (MKS Instruments Incorporated)
• Original Assignee: Electro Scientific Industries Incorporated (MKS Instruments Incorporated)
• Inventors: Wilson, Stephen S.
• Primary Examiner(s): Ahmed, Samir
• Assistant Examiner(s): BHATNAGAR, ANAND P
• Application Number:
  US09/366,905
• Time in Patent Office: 1,812 Days
• Field of Search: 382/209, 382/155-160, 382/199-200, 382/217, 382/103, 382/218, 382/219, 382/151, 348/129, 348/130, 348/91
• US Class Current: 382/217
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 10/443 by matching or filtering

Abstract

The present invention solves the problem of repeated operator intervention in a vision system by providing multiple templates which may be used to locate an object within an image.

Claims

1. A method of locating a first object within a series of images comprising:

providing a first template representative of the first object;

applying the first template to the first image within a series of images;

reporting the location of the first object if the first object is found within the first image;

training an additional template from the first image if the first object is not located in the first image by the first template, the additional template being characteristic of the first object, the first and additional template forming a collection of templates;

applying the first template to a subsequent image to find the first object;

determining if an additional template is available from the collection of templates if the first template does not find the first object in the subsequent image, and;

applying the additional template to the subsequent image to find the first object.

2. A method as in claim 1 wherein the template is a kernel applied by binary vector correlation.

3. A method as in claim 1 wherein the kernel is trained using simulated annealing.

4. A method as in claim 1 wherein the kernel represents pairs of points straddling an edge and the kernel is applied using gray scale vector correlation.

5. A method as in claim 1 wherein the additional template is trained in response to an operator assist call.

6. A method as in claim 1 wherein additional template is trained without manual assistance.

7. A method as in claim 1 wherein only one additional template is trained for each operator assist.