Methods and apparatuses for detecting classifying and measuring spot defects in an image of an object

• US 7,162,073 B1
• Filed: 11/30/2001
• Issued: 01/09/2007
• Est. Priority Date: 11/30/2001
• Status: Expired due to Fees

• Current Assignee: Cognex Technology & Investment Corporation (Cognex Corporation)
• Original Assignee: Cognex Technology & Investment Corporation (Cognex Corporation)
• Inventors: Koljonen, Juha, Davis, Jason, Wagman, Adam, Morje, Prabhav, Bachelder, Ivan, Akgul, Yusuf
• Primary Examiner(s): Wu; Jingge
• Assistant Examiner(s): KIM, CHONG R
• Application Number:
  US10/001,177
• Time in Patent Office: 1,866 Days
• Field of Search: 382141-152, 382/190, 382/199, 382/204, 382/212, 382/213, 356/73.1, 356/237.1, 356/239.2
• US Class Current: 382/149
• CPC Class Codes: G01N 21/8851 Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges G01N21/8806 and G01N21/93 - G01N21/95692 take precedence; optical measurement of dimensions G01B11/00; optical scanning G02B26/10; image transformation G06T3/00; computerised image enhancement G06T5/00; image processing per se for flaw detection G06T7/0002
• CPC Class Codes: G06T 2207/10056 Microscopic image
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 7/0004 Industrial image inspection
• CPC Class Codes: G06T 7/174 involving the use of two or more images
• CPC Class Codes: G06T 7/194 involving foreground-background segmentation
• CPC Class Codes: G06V 10/28 Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
• CPC Class Codes: G06V 2201/06 Recognition of objects for industrial automation

Abstract

A method is provided for detecting spot defects on an object when an allowable variation (called the “background”) in the appearance of the object can be modeled. Methods are also provided for measuring and classifying detected spot defects. An alignment model is used to align the image of the object, a background model is then used to estimate the (possibly different) background in each region, and each background is substantially removed from the image so as to form a foreground image on which blob analysis can be applied to detect spot defects, the blob analysis using a threshold image that accommodates different noise statistics for each region. The method facilitates robust spot defect inspection of fiber optic end faces, or of any object with different object regions. The method also allows use of blob analysis over a larger range of conditions, including conditions that make simple blob analysis infeasible.

Claims

1. A method for detecting spot defects in an image of an object, the method comprising:

acquiring an image of the object at only a single plane of focus;

aligning an alignment model with at least one object region within the image so as to provide at least one aligned object region;

estimating, using the image, a background image representing allowable and expected appearance within the at least one aligned object region, by using a median filter of a larger spatial scale to provide a mask for use with a median filter of a smaller spatial scale;

removing the background image from the image within the at least one aligned object region so as to provide a foreground image having at least one foreground region;

computing a threshold image for at least one of the foreground regions in the foreground image;

applying the threshold image to the foreground image within at least one of the foreground regions so as to provide a defect image; and

using connected component analysis to form candidate defects in the defect image.

2. The method of claim 1, further comprising:

classifying candidate defects as one of non-defects, and one of a plurality of possible defect types.

3. The method of claim 1, further comprising:

measuring at least one metric for each candidate defect.

4. The method of claim 3, wherein measuring at least one metric for each defect includes:

measuring sub-pixel boundaries of blobs in the foreground image by edge detection along the boundaries of blobs.

5. The method of claim 3, wherein measuring at least one metric for each defect includes:

measuring sub-pixel boundaries of blobs in the foreground image by tracking along the boundaries of blobs.

6. The method of claim 3, wherein measuring at least one metric for each defect includes:

measuring sub-pixel boundaries of blobs in the foreground image by performing sub-pixel interpolation along the boundaries of blobs.

7. The method of claim 1, wherein aligning performs deformable alignment.

8. The method of claim 7, wherein deformable alignment uses core alignment.

9. The method of claim 1, wherein the object is a fiber-optic end face.

10. The method of claim 1, wherein aligning is performed using a rotation-invariant and scale-invariant alignment method.

11. The method of claim 1, wherein the median filter is an annular median filter.

12. The method of claim 1, wherein the median filter of the larger spatial scale is thresholded to provide the mask.

13. The method of claim 1, wherein estimating includes:

iteratively fitting a smooth surface to the image within the at least one aligned object region.

14. The method of claim 13, wherein the smooth surface is a polynomial curve.

15. The method of claim 13, wherein outliers are removed after each iterative fitting.

16. The method of claim 1, wherein removing includes:

subtracting the background image from the image.

17. The method of claim 1, wherein computing a threshold image for at least one of the foreground regions in the foreground image includes:

removing from the at least one of the foreground regions all pixels with a gray level above a threshold.

18. The method of claim 1, wherein computing a threshold image for at least one of the foreground regions in the foreground image includes:

using histogram analysis.

19. The method of claim 1, wherein computing a threshold image for at least one of the foreground regions in the foreground image includes:

removing, from the at least one of the foreground regions in the foreground image, pixels with a gray level between two prominent peaks of a mufti-modal distribution.

20. The method of claim 1, wherein applying the threshold image to the foreground image within at least one of the foreground regions so as to provide a defect image includes:

subtracting the threshold image from the foreground image within at least one of the foreground regions so as to provide a defect image.

21. A method for detecting spot defects in an image of an object, the method comprising:

acquiring an image of the object at only a single plane of focus;

aligning an alignment model with at least one object region within the image so as to provide at least one aligned object region;

estimating, using the image, a background image representing allowable and expected appearance within the at least one aligned object region;

removing the background image from the image within the at least one aligned object region so as to provide a foreground image having at least one foreground region;

computing a threshold image for at least one of the foreground regions in the foreground image by adding the gradient of the background image to the threshold image;

applying the threshold image to the foreground image within at least one of the foreground regions so as to provide a defect image; and

using connected component analysis to form candidate defects in the defect image.