Second opinion method of pattern recognition error reduction

• US 4,987,604 A
• Filed: 06/02/1989
• Issued: 01/22/1991
• Est. Priority Date: 06/02/1989
• Status: Expired due to Term

• Current Assignee: Delphi Technologies, Inc. (BorgWarner Incorporated)
• Original Assignee: Delco Electronics Corporation (Motors Liquidation Co. GUC Trust)
• Inventors: Rouch, Stephen F.
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Johns, Andrew W.
• Application Number:
  US07/360,680
• Time in Patent Office: 599 Days
• Field of Search: 382/8, 382/62, 382/37, 382/38, 382/30, 382/34
• US Class Current: 382/151
• CPC Class Codes: G06V 10/7515 Shifting the patterns to accommodate for positional errors

Abstract

A method of operating optical pattern recognition apparatus which relies upon second opinions in which stored offset coordinates match one another within a predetermined tolerance to avoid unnecessarily high numbers of rejects, undesirable levels of risk in terms of pattern misalignments, or excessive cycle times in terms of unnecessary looping through different grey level thresholds.

Claims

1. A method of operating a pattern recognition system which digitizes the image of a work item, using a selected threshold to determine a grey level to be digitized as black instead of white, compares the pixels of the digitized work item image with those of a digitized reference image stored in memory, and produces both the offset coordinates of the work item image relative to the reference image and a quality number representing the match between the pixels of the digitized work item image and those of the digitized reference image, said method comprising the steps of:

seeking a quality number on the basis of one grey level threshold;

accepting the offset coordinates associated with said quality number if said quality number exceeds a first quality reference level;

otherwise seeking another quality number on the basis of another grey level threshold;

accepting the offset coordinates associated with said other quality number if said other quality number equals or exceeds said first quality reference level;

otherwise comparing both of said quality numbers with a second lower quality reference level; and

accepting the offset coordinates of one of said quality numbers if the offset coordinates of both quality numbers match one another within a predetermined tolerance.

2. A method of operating a pattern recognition system which digitizes the image of a work item, using a predetermined threshold to determine a grey level to be digitized as black instead of white, compares the pixels of the digitized work item image with those of a digitized reference image stored in memory, and produces both the offset coordinates of the work item image and a quality number representing the match between the pixels of the digitized work item image relative to the reference image and those of the digitized reference image, said method comprising the steps of:

selecting a first quality reference level and at least one lower quality reference level;

generating a first quality number and corresponding offset coordinates;

comparing said first quality number with said first reference level;

accepting associated offset coordinates if said first quality number equals or exceeds said first reference level;

generating at least one additional quality number and corresponding offset coordinates on the basis of a corresponding different grey level threshold if said first quality number does not equal or exceed said first reference level;

accepting its associated offset coordinates if said additional quality number equals or exceeds said first reference level;

otherwise storing the two highest quality numbers otherwise generated and their corresponding offset coordinates in memory;

comparing the stored quality numbers with said lower reference level;

comparing the stored sets of offset coordinates with one another if both of said stored quality numbers equal or exceed said lower reference level; and

accepting the offset coordinates of one of the stored quality numbers if those associated with both stored quality numbers match one another within a predetermined tolerance.

3. A method of operating a pattern recognition system which digitizes the image of a work item, using a predetermined threshold to determine a grey level to be digitized as black instead of white, compares the pixels of the digitized work item image with those of a digitized reference image stored in memory, and produces both the offset coordinates of the work item image relative to the reference image and a quality number representing the match between the pixels of the digitized work item image and those of the digitized reference image, said method comprising the steps of:

selecting a first quality reference level and second and third successively lower quality reference levels;

generating a first quality number and corresponding offset coordinates;

comparing said first quality number with said first reference level;

accepting its associated offset coordinates if said first quality number equals or exceeds said first reference level;

generating at least one additional quality number and corresponding offset coordinates on the basis of a different grey level threshold if said first quality number does not equal or exceed said first reference level;

accepting its associated offset coordinates if said additional quality number equals or exceeds said first reference level;

otherwise storing the two highest quality numbers generated and their corresponding offset coordinates in memory;

comparing the stored quality numbers with said second reference level;

comparing the stored sets of offset coordinates with one another if both of said stored quality numbers equal or exceed said second reference level;

accepting the offset coordinates associated with one of the stored quality numbers if those of both stored quality numbers match one another within a predetermined tolerance;

comparing the stored quality numbers with said third reference level if at least one is less than said second reference level;

comparing the stored sets of offset coordinates with one another if both of said stored quality numbers equal or exceed said third reference level; and

accepting the offset coordinates associated with one of the stored quality numbers if those of both stored quality numbers match one another within a predetermined tolerance.

4. A method of locating the position of a work item comprising the steps of:

forming a digitized image of the work item, using a first threshold T1 to determine a grey level to be digitized black instead of white;

comparing the image of the work item with a stored reference image;

generating both a first quality number Q1 representing the match between the image of the work item and the reference image and a first set of corresponding location coordinates (X1,Y1) for the work item;

comparing Q1 with a first predetermined quality reference level L1;

accepting (X1,Y1) as correct if Q1 is equal to or greater than L1;

otherwise forming a second digitized image of the work item using a second grey level threshold T2;

comparing said second image with said stored reference image;

generating both a second quality number Q2 representing the match between the second image of the work item and said reference image and a second set of location coordinates (X2,Y2) for the work item;

accepting (X2,Y2) as correct if Q2 is equal to or greater than L1;

otherwise comparing both Q1 and Q2 with a second lower quality reference level L2;

comparing (X1,Y1) and (X2,Y2) with one another if both Q1 and Q2 are equal to or greater than L2; and

accepting either (X1,Y1) or (X2,Y2) as correct if both match one another within a predetermined tolerance.

5. A method of locating the position of a work item comprising the steps of:

forming a digitized image of the work item, using a first threshold T1 to determine a grey level to be digitized as black instead of white;

comparing the image of the work item with a stored reference image;

generating both a first quality number Q1 representing the match between the image of the work item and the reference image and a set of corresponding location coordinates (X1,Y1) for the work item;

storing both Q1 and (X1,Y1) in memory;

selecting a first quality reference level L1 and a second lower quality reference level L2;

comparing Q1 with L1;

accepting (X1,Y1) as correct if Q1 is equal to or greater than L1;

otherwise forming a second digitized image of the work item using a second grey level threshold T2;

comparing said second image with said stored reference image;

generating both a second quality number Q2 representing the match between the second image of the work item and said reference image and a second set of location coordinates (X2,Y2) for the work item;

storing both Q2 and (X2,Y2) in memory;

accepting (X2,Y2) as correct if Q2 is equal to or greater than L1;

otherwise comparing both Q1 and Q2 with L2;

comparing (X1,Y1) and (X2,Y2) with one another if both Q1 and Q2 are greater than L2; and

accepting one of (X1,Y1) and (X2,Y2) as correct if said sets match one another within a predetermined tolerance.

6. A method of locating the position of a work item comprising the steps of:

forming a digitized image of the work item, using a first threshold T1 to determine a grey level to be digitized as black instead of white;

comparing the image of the work item with a stored reference image;

generating both a quality number Q1 representing the match between the image of the work item and the reference image and a set of corresponding location coordinates (X1,Y1) for the work item;

storing both Q1 and (X1,Y1) in memory;

selecting first, second, and third quality reference levels L1, L2, and L3, each successively lower than the other;

comparing Q1 with L1;

accepting (X1,Y1) as correct if Q1 is equal to or greater than L1;

otherwise forming a second digitized image of the work item using a second grey level threshold T2;

comparing said second image with said stored reference image;

generating both a second quality number Q2 representing the match between the second image of the work item and said reference image and a second set of location coordinates (X2,Y2) for the work item;

storing both Q2 and (X2,Y2) in memory;

accepting (X2,Y2) as correct if Q2 is equal to or greater than L1;

otherwise comparing both Q1 and Q2 with L2;

comparing (X1,Y1) and (X2,Y2) with one another if both Q1 and Q2 are equal to or greater than L2;

accepting either (X1,Y1) or (X2,Y2) as correct if both match one another within a predetermined tolerance;

otherwise comparing both of Q1 and Q2 with L3;

comparing (X1,Y1) and (X2,Y2) with one another if both Q1 and Q2 are greater than L3; and

accepting either (X1,Y1) or (X2,Y2) as correct if both match one another within a predetermined tolerance.

7. The method of locating the position of a work item of claim 6 further comprising the steps of:

forming a third digitized image of the work item using a third grey level threshold T3;

comparing said third image with said stored reference image;

generating both a third quality number Q3 representing the match between the third image of the work item and said reference image and a third set of location coordinates (X3,Y3) for the work item;

substituting Q3 and (X3,Y3) in memory for the lesser of the two stored quality numbers and its associated set of location coordinates;

accepting (x3,Y3) as correct if Q3 is equal to or greater than L1;

comparing both of the quality numbers stored in memory with L2;

comparing the sets of location coordinates stored in memory with one another if both of said stored quality numbers are equal to or greater than L2;

accepting one of the stored sets of location coordinates as correct if said sets match one another within a predetermined tolerance;

otherwise comparing both of the quality numbers stored in memory with L3;

comparing the sets of location coordinates stored in memory with one another if both of said stored quality numbers are greater than L3; and

accepting one of the sets of location coordinates stored in memory as correct if said sets match one another within a predetermined tolerance.