Method of and device for inspecting images to detect defects

• US 6,804,381 B2
• Filed: 04/18/2001
• Issued: 10/12/2004
• Est. Priority Date: 04/18/2000
• Status: Expired due to Term

• Current Assignee: City University of Hong Kong
• Original Assignee: University of Hong Kong (Government of Hong Kong Special Administrative Region)
• Inventors: Kumar, Ajay, Pang, Kwok-Hung Grantham
• Primary Examiner(s): Johns, Andrew W.
• Application Number:
  US09/837,065
• Publication Number:
  US 20020054293A1
• Time in Patent Office: 1,273 Days
• Field of Search: 382/111, 382/141, 382/173, 382/260, 382/264, 382/270, 348/88, 348/92, 348/125, 348/128, 356/430, 356/431, 356/238.1, 356/238.2, 356/238.3, 162/198
• US Class Current: 382/111
• CPC Class Codes: G01N 21/898 Irregularities in textured or patterned surfaces, e.g. textiles, wood

Abstract

The present invention relates to a method for automated defect detection in textured materials. The present invention utilizes linear Finite Impulse Response (FIR) filters with optimized energy separation. Specifically, the invention provides a method of inspecting industrial products for defects. The method has steps of: automated design of optimized filters from samples of products, using these optimal filters to filter the acquired images of product under inspection, computing the energy of each pixel in a local region, and finally segmenting the defect by thresholding each pixel. The present invention also relates to a method of inspection of unknown (unsupervised) defects in web materials. In an unsupervised inspection, information from a finite number of optimal filters is combined using a data fusion module. This module attempts to nullify the false alarm associated with the information arriving from different channels.

Claims

1. A method of inspecting a web material containing defects, the method comprising:

acquiring an image of the web material, the acquired image comprising a plurality of pixels;

attenuating the pixels in the acquired image corresponding to a defect free region;

obtaining the energy of each pixel; and

segmenting the defects.

2. The method of claim 1, wherein the acquired image is filtered.

3. The method of claim 1, wherein an optimal filter is used to attenuate the pixels in the acquired image corresponding to a defect free region.

4. The method of claim 1, wherein a nonlinear function is used to obtain the energy of each pixel.

5. The method according to claim 4, wherein the nonlinear function is squaring and absolute values of each filtered pixel.

6. The method of claim 1 further comprising obtaining a local energy estimation for each pixel in a local region.

7. The method of claim 6, wherein a smoothing filter is used to compute the local energy estimation of each pixel in a local region around the pixel.

8. The method of claim 1, wherein the acquired images are thresholded to segment the defects.

9. The method of claim 1, wherein the web material under inspection is selected from the group consisting of paper, fabric, plastic wood, and steel.

10. A method of inspecting unsupervised web materials containing defects, the method comprising:

designing a plurality of optimal filters for different categories of defects;

combining the output from each of the optimal filters in a data fusion module to obtain a fused output image; and

segmenting defects in the fused output image.

11. The method of claim 10, wherein defects in the fused output image are segmented by thresholding.

12. The method of claim 10, wherein the web material comprises a woven material.

13. The method of claim 10, wherein two filters for horizontal and vertical directions are used.

14. The method of claim 10, wherein the web material under inspection is selected from the group consisting of paper, fabric, plastic, wood, and steel.

15. A method of inspecting defects in textured materials for industrial automation, the method comprising:

imaging the material under inspection;

using pre-designed optimal filters to attenuate pixels in acquired images corresponding to a defect free region;

using a nonlinear function to compute energy of each pixel;

using a smoothing filter to compute energy of every pixel in a local region; and

thresholding the acquired images to segment the defects.

16. The method of claim 15, wherein the pre-designed optimal filters include at least one filter device for detecting a fabric defects, the filter comprising an optimal object function of:

$J_1<br><br>\left(h_{\mathrm{op}}\right)=\frac{{\mathrm{\mu <br><br>}}_{f_d}}{{\mathrm{\mu <br><br>}}_{f_r}}=\frac{h_{\mathrm{op}}^T<br><br>R_{{\mathrm{ii}}_d}<br><br>h_{\mathrm{op}}}{h_{\mathrm{op}}^T<br><br>R_{{\mathrm{ii}}_r}<br><br>h_{\mathrm{op}}}$

wherein μ

<sub>ƒ

d</sub> and μ

<sub>ƒ

r</sub> designate the average local energy estimate for a fabric with defect and without defect, respectively; and

wherein Rii=E{i(x,y)i^τ

(x,y)} is the L×

L(L=M×

N) correlation matrix and can be constructed from the autocorrelation function of the image i(x,y).

17. The method of claim 15, wherein the pre-designed optimal filters include at least one filter device for detecting fabric defects, the filter being designed to optimize the object function of:

$J_2<br><br>\left(h_{\mathrm{op}}\right)=\frac{{\left({\mathrm{\mu <br><br>}}_{f_d}-{\mathrm{\mu <br><br>}}_{f_r}\right)}^2}{{\mathrm{\mu <br><br>}}_{f_d}<br><br>{\mathrm{\mu <br><br>}}_{f_r}}$

wherein μ

<sub>ƒ

d</sub> and μ

<sub>ƒ

r</sub> designate the average local energy estimate for a fabric with defect and without defect, respectively.

18. The method of claim 15, wherein the pre-designed optimal filters include at least one filter device for detecting fabric defects, the filter being designed to optimize the object function of:

$J_3<br><br>\left(h_{\mathrm{op}}\right)=\frac{{\left({\mathrm{\mu <br><br>}}_{f_d}-{\mathrm{\mu <br><br>}}_{f_r}\right)}^2}{{\mathrm{\sigma <br><br>}}_{f_d^2}+{\mathrm{\sigma <br><br>}}_{f_r^2}}$

wherein μ

<sub>ƒ

d</sub> and μ

<sub>ƒ

r</sub> designate the average local energy estimate for a fabric with defect and without defect, respectively; and

σ

<sub>ƒ

d</sub> and σ

<sub>ƒ

r</sub> designate variances of local energy estimate.