Fast density estimation method for defect inspection application

Patent metadata

Field	Value
Publication number	US-9846929-B2
Application number	US-201615079062-A
Country	US
Kind code	B2
Filing date	Mar 24, 2016
Priority date	Mar 24, 2016
Publication date	Dec 19, 2017
Grant date	Dec 19, 2017

Abstract

The present invention provide a high speed spatial density estimation algorithm to estimate defect density maps for blob analysis in an image processing field for inspection. The method of the present invention uses a rotated L1 epsilon-ball neighborhood mask for determining a defect density for each of target pixels to generate a defect density map for defect detection of an object. The present method is capable of providing high detection speed and substantially eliminating influence of the noise from images.

First claim

What is claimed is:

1. A method for inspecting one or more defects of an object by using defect density estimation, the method comprising:
capturing one or more images of the object;
estimating defect response from the one or more captured images;
generating, from the defect response estimated from the one or more images of the object, a binary matrix comprising one or more defect elements;
determining a first neighborhood mask for each target pixel of the binary matrix, wherein a first neighborhood mask for a respective target pixel of the binary matrix is determined based on a predetermined neighborhood distance with respect to the respective target pixel;
determining a second neighborhood mask for each target pixel of the binary matrix, wherein a second neighborhood mask for a respective target pixel of the binary matrix is determined by rotating the first neighborhood mask for the respective target pixel through a predetermined angle;
determining a density of defect elements of the binary matrix within a second neighborhood mask corresponding to each target pixel of the binary matrix; and
determining the one or more defects of the object based on the determined density of defect elements of the binary matrix for each target pixel.

2. The method of claim 1 , wherein each of the one or more captured images is a 2-dimension gray intensity image, a color image, a depth image, or a binary image.

3. The method of claim 1 , wherein the binary matrix is generated by a thresholding method.

4. The method of claim 3 , wherein the thresholding method includes a method, which converts a matrix to a binary matrix.

5. The method of claim 3 , wherein the thresholding method includes an adaptive thresholding method or a traditional thresholding method.

6. The method of claim 1 , wherein each of the one or more defect elements is a non-zero element.

7. The method of claim 1 , wherein each of the one or more defect elements is a zero element.

8. The method of claim 1 , wherein the first neighborhood mask is a first L1 epsilon-ball neighborhood mask.

9. The method of claim 1 , wherein the predetermined neighborhood distance is a Manhattan distance between the respective target pixel and any point at an edge of the first neighborhood mask for the respective target pixel.

10. The method of claim 1 , wherein the predetermined neighborhood distance is an optimal neighborhood radius calculated based on input defect image data of the object and a prediction model.

11. The method of claim 10 , wherein the prediction model is determined by conducting a data-fusion process and a regression based learning process.

12. The method of claim 1 , wherein the predetermined neighborhood distance is an optimal neighborhood radius determined through a machine learning based process.

13. The method of claim 12 , wherein the machine learning based process comprises:
conducting a data-fusion process and a regression based learning process to generate a prediction model; and
calculating the optimal neighborhood radius based on data of the one or more captured images and the prediction model.

14. The method of claim 1 , wherein the predetermined angle is an angle of 45 degree, 135 degree, 225 degree, or 315 degree.

15. The method of claim 1 , wherein the density of defect elements of the binary matrix is determined by counting the defect elements of the binary matrix.

16. The method of claim 1 , further comprising:
generating a defect density map based on the determined density of defect elements of the binary matric for determining the defects of the object.

17. The method of claim 16 , further comprising:
classifying noise of the one or more captured images from the defect density map to output a defect image.

18. The method of claim 1 , further comprising:
classifying defect pixels and noise pixels from the one or more captured images based on the determined density of defect elements of the binary matrix to output a defect image; and
post-processing the defect image with one or more morphological operations or blob filtering in terms of a geometric size, an area of connected pixels, or an aspect to reduce the noise pixels.

19. The method of claim 18 , wherein the one or more morphological operations include eroding, dilating, opening, and/or closing in digital image processing context.

20. The method of claim 18 , wherein the step of classifying the defect pixels and the noise pixels is done with one or more defect density images based on the determined density of defect elements of the binary matrix.

Assignees

• Hong Kong Applied Science & Tech Research Inst Co Ltd

Inventors

• Zou Weiwen

Classifications

• G06T7/0004 Primary

  Industrial image inspection · CPC title

• G06T7/0008 Primary

  checking presence/absence · CPC title

• G06T5/30

  Erosion or dilatation, e.g. thinning · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/20036

  Morphological image processing · CPC title