초록 ▼

A defect detection method and system for the automated visual inspection of web materials is provided. The invention utilize real Gabor function (RGF) filters with a non-linear function for estimating the local energy. An example method for quality assurance using automated visual inspection in acco

A defect detection method and system for the automated visual inspection of web materials is provided. The invention utilize real Gabor function (RGF) filters with a non-linear function for estimating the local energy. An example method for quality assurance using automated visual inspection in accordance with the invention includes the steps of: automated design of a bank of RGF filters; using these RGF filters to sample the features of image under inspection; using a non-linear function to compute local energy estimate in the filtered images; combining all the filtered images using image fusion; and finally thresholding the resultant image to segment the defects in the inspection image. Possible embodiments of the described invention include detection of only a class of defects using a single tuned real Gabor filter or a bank of real Gabor functions.

대표청구항 ▼

1. A quality assurance method for inspecting a surface of an object for defects against a texture background in uniform and textured web products using automated visual inspection comprising the steps of:imaging a web material to obtain an image;detecting, automatically, at least one frequency compo

1. A quality assurance method for inspecting a surface of an object for defects against a texture background in uniform and textured web products using automated visual inspection comprising the steps of:imaging a web material to obtain an image;detecting, automatically, at least one frequency component associated with the background texture of a reference image;applying a pre-designed set of real Gabor function based filters associated with the at least one frequency component and a mask parameter to evaluate features of the image under inspection and generate a plurality of derived images corresponding to the pre-designed set of real Gabor function based filters;estimating energy in a local region by computing a local energy function from at least one of the plurality of derived images to generate a blob-detecting image;combining at least two blob-detecting images with the aid of an image fusion rule to obtain a resultant image; andthresholding the resultant image to further segment any possible defect(s). 2. The method according to claim 1 wherein the combining step employs image using Bernoulli's rule of combination. 3. The method according to claim 1 wherein the combining step further employs image fusion using vector addition of pixels. 4. The method according to claim 1 wherein the step of imaging includes generating a set of pixels corrected for illumination irregularities. 5. The method of claim 1 wherein the local energy function is a rectified sigmoid in the step of estimating energy. 6. The method of claim 5 wherein the rectified sigmoid is tuned by adjusting β in the range 0.2-0.3. 7. The method according to claim 1, wherein image correction compensates for illumination irregularities by comparison with a reference. 8. The method according to claim 1, further including the step of computing the threshold by estimating at least one maximum value for a pixel in an integrated image generated from the reference image. 9. The method according to claim 1 wherein the surface under inspection is that of a textile. 10. The method according to claim 1 wherein the surface under inspection is that of wood. 11. The method according to claim 1 wherein the surface under inspection is that of a metal. 12. The method according to claim 1 wherein the surface under inspection is that of a paper. 13. The method according to claim 1 wherein the mask parameter is 5×5. 14. The method according to claim 1, wherein the mask parameter is 9×9. 15. The method according to claim 1, wherein the mask parameter is 7×7. 16. A method for quality assurance in uniform and textured web products with known yarn density using automated visual inspection comprising the steps of:imaging a web material to obtain an image of a surface;applying a pre-designed Gabor filter to the image to generate a filtered image;applying a nonlinear function to a pixel in the filtered image to generate a resultant image; andthresholding the resultant image to segment at least one defect. 17. The method according to claim 16, wherein the nonlinear function is a median filter. 18. The method according to claim 16, wherein the nonlinear function is a squaring function. 19. The method according to claim 16, wherein computing the threshold includes estimating a maximum value for a pixel in an integrated image generated from a defect-free web material. 20. The method according to claim 16, wherein the surface is a textile surface. 21. The method according to claim 16, wherein the surface is a wood surface. 22. The method according to claim 16, wherein the surface is a metal surface. 23. The method according to claim 16, wherein the surface is a paper surface. 24. A method for quality assurance in uniform and textured web products with known yarn density using automated visual inspection comprising the steps of:imaging a defect-free web material by generating a set of pixels corrected for illumination irregularities to obtain an image;selecting at least one frequency component associated with the background texture of the image;designing at least one real Gabor function based filter using the at least one frequency component associated with the background texture of the image; andcomputing, automatically, a threshold associated with the at least one real Gabor function based filter using the image. 25. The method according to claim 24 wherein image correction compensates for illumination irregularities by comparison with a reference. 26. The method according to claim 24 wherein the step of selecting includes automatically determining the at least one frequency component associated with the background texture of the image from a set of pixels in the warp direction whereby reducing the need for manual tuning. 27. The method according to claim 24 wherein the mask parameter is 5×5. 28. The method according to claim 24 wherein the mask parameter is 9×9. 29. The method according to claim 24 wherein the mask parameter is 7×7. 30. The method according to claim 24 wherein the surface under inspection is that of a textile. 31. A method for quality assurance in the uniform and textured web products using automated visual inspection comprising the steps of:imaging a defect-free web material by generating a set of pixels corrected for illumination irregularities to obtain an image;detecting, automatically, a first frequency component associated with the background texture of the image from a set of pixels in the warp direction and a second frequency component associated with the background texture of the image from a set of pixels in the weft direction;selecting a highest frequency component following comparison of at least the first and the second frequency components;designing at least one real Gabor function based filter using the selected frequency component associated with the background texture of the image; andcomputing, automatically, a threshold associated with the at least one real Gabor function based filter using the image. 32. The method according to claim 31, wherein image correction compensates for illumination irregularities by comparison with a reference. 33. The method according to claim 31, wherein the surface under inspection is that of a textile. 34. The method according to claim 31, wherein the step of computing the threshold includes estimating a maximum value for a pixel in an integrated image generated from the defect-free web material. 35. The method of claim 34 wherein the integrated image is generated from a plurality of real Gabor functions that include the at least one real Gabor function. 36. A system for automatically detecting defects against a textured background in a surface of a material, the system comprising:image acquisition means;at least one processor configured to execute the steps for detecting, automatically, at least one frequency component associated with the background texture of a reference image, applying a pre-designed set of real Gabor function based filters associated with the at least one frequency component and a mask parameter to evaluate features of the image under inspection and generate a plurality of derived images corresponding to the pre-designed set of real Gabor function based filters, estimating energy in a local region by computing a local energy function from at least one of the plurality of derived images to generate a blob-detecting image, combining at least two blob-detecting images with the aid of an image fusion rule to obtain a resultant image, and thresholding the resultant image to further segment any possible defect(s); anda memory for storing at least one mean and standard deviation value for a blob detecting image, and at least one threshold based on a defect-free image.