Product appearance inspection methods and apparatus employing low variance filter

• US 5,237,621 A
• Filed: 08/08/1991
• Issued: 08/17/1993
• Est. Priority Date: 08/08/1991
• Status: Expired due to Term

• Current Assignee: Philip Morris USA Incorporated (Altria Group, Inc.)
• Original Assignee: Philip Morris USA Incorporated (Altria Group, Inc.)
• Inventors: Dante, Henry M., Cox, Kenneth A., Maher, Robert J.
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Prikockis, Larry J.
• Application Number:
  US07/742,323
• Time in Patent Office: 740 Days
• Field of Search: 382/14, 382/27, 382/22, 382/55, 382/8, 382/34
• US Class Current: 382/143
• CPC Class Codes: G06T 2207/20036 Morphological image processing
• CPC Class Codes: G06T 2207/30144 Printing quality
• CPC Class Codes: G06T 2207/30164 Workpiece; Machine component
• CPC Class Codes: G06T 7/001 using an image reference approach

Abstract

Images such as product images are identified as substantially similar to one or more reference images by finding pixels which have substantially the same value or values in all or substantially all of the reference images. Image erosion and/or dilation may be used in processing the reference image data to help identify pixels which can always be expected to have the same value in all acceptable images. The values associated with the corresponding pixels in a product image are combined and compared to an expected value, and the dot product between the product image and the discriminant function is equal to a predetermined value. The values associated with pixels which correspond to regions not similar in substantially all of the reference images are combined to produce a discriminant function.

Claims

1. The method of determining whether a product image is substantially similar to a plurality of training images comprising the steps of:

similarly subdividing each of said training images and said product image into a plurality of pixel regions;

associating a numerical value with each of said pixel regions in each of said training images and said product image, each of said numerical values being indicative of a predetermined image characteristic of the associated pixel region in the associated training image or product image;

identifying the pixel regions in substantially all of said training images which have substantially the same associated numerical value;

identifying the pixel regions in said product image which correspond to the pixel regions identified in the preceding step;

combining the numerical values associated with the pixel regions in said product image identified in the preceding step to produce a composite value;

combining the numerical values associated with the pixel regions in said training images not identified as having substantially the same value in substantially all of said training images to produce a discriminant value associated with each of said not identified pixel regions;

computing the dot product of said discriminant values and the numerical values associated with the pixel regions in said product image which correspond to said not identified pixel regions in said training images; and

identifying said product image as substantially similar to said training images only if said composite value is substantially equal to a predetermined composite value and said dot product is substantially equal to a predetermined dot product value.

2. The method defined in claim 1 wherein said step of associating a numerical value comprises, for each of said training images, the step of:

processing said image so that the numerical value associated with each pixel region has a first of two predetermined values if said predetermined image characteristic of said pixel region has a predetermined relationship to a predetermined threshold image characteristic, and so that the numerical value associated with each pixel region has a second of said two predetermined values if said predetermined image characteristic does not have said predetermined relationship to said predetermined threshold image characteristic.

3. The method defined in claim 2 wherein said step of identifying the pixel regions in substantially all of said training images comprises the step of:

identifying the pixel regions which have said first of said two predetermined values in substantially all of said training images.

4. The method defined in claim 2 wherein said step of identifying the pixel regions in substantially all of said training images comprises the step of:

identifying the pixel regions which have said first of said two predetermined values in substantially all of said training images, and the pixel regions which have said second of said two predetermined values in substantially all of said training images.

5. The method defined in claim 1 further comprising the step of:

edge enhancing each training image prior to performing said step of associating a numerical value.

6. The method defined in claim 1 wherein said step of associating a numerical value comprises, for each of said training images, the steps of:

edge enhancing said training image; and

processing said edge enhanced training image so that numerical value associated with each pixel region has a first of two predetermined values if said predetermined image characteristic of said pixel region has a predetermined relationship to a predetermined threshold image characteristic, and so that the numerical value associated with each pixel region has a second of said two predetermined values if said predetermined image characteristic does not have said predetermined relationship to said predetermined threshold image characteristic.

7. The method defined in claim 6 wherein said step of identifying the pixel regions in substantially all of said training images comprises the step of:

identifying the pixel regions which have said first of said two predetermined values in substantially all of said edge enhanced training images.

8. The method defined in claim 1 wherein said step of combining the numerical values comprises the step of:

adding together the numerical values associated with the pixel regions in said product image identified in the preceding step to produce said composite value.

9. The method defined in claim 1 wherein said step of combining the numerical values comprises the steps of:

associating a filter value with each pixel region identified as having substantially the same numerical value in substantially all of said training images; and

computing the dot product of said filter values and said numerical values associated with the corresponding pixel regions in said product image to produce said composite value.

10. The method defined in claim 1 wherein said step of identifying the pixel regions in substantially all of said training images comprises the steps of:

for each pixel region, combining said numerical values for all of said training images to produce a combined value associated with each pixel region; and

identifying the pixel regions having a substantially common combined value.

11. The method defined in claim 10 further comprising the step of:

after said step of combining said numerical values for all of said training images and before said step of identifying the pixel regions having a substantially common combined value, subjecting each of said combined values to an erosion operation which changes said combined value to a value other than said common combined value unless said combined value and a predetermined number of adjacent combined values have said common combined value.

12. The method defined in claim 10 further comprising the step of:

after said step of combining said numerical values for all of said training images and before said step of identifying the pixel regions having a substantially common combined value, subjecting each of said combined values to a dilation operation which changes a predetermined number of combined values adjacent to each combined value having said common combined value to said common combined value.

13. The method defined in claim 11 further comprising the step of:

after said subjecting step and before said step of identifying the pixel regions having a substantially common combined value, subjecting each of said combined values to a dilation operation which changes a predetermined number of combined values adjacent to each combined value having said common combined value to said common combined value.