Optics: measuring and testing – By monitoring of webs or thread – For flaws or imperfections
Reexamination Certificate
2001-10-25
2004-06-22
Font, Frank G. (Department: 2877)
Optics: measuring and testing
By monitoring of webs or thread
For flaws or imperfections
C356S238100, C356S238200
Reexamination Certificate
active
06753965
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a method and system for automated visual monitoring of surface characteristics, and more particularly to automated optical inspection of textile characteristics to detect fabric defects and anomalies.
BACKGROUND OF THE INVENTION
The development of a reliable automated visual inspection system for quality assurance requires extensive research and development effort. Human inspectors typically perform visual inspection for quality assurance in industrial products. The disadvantage with the manual inspection are: (1) low speed, (2) high cost, (3) inability to perform real-time inspection and, (4) the serious limitations on the range of detectable defects. Therefore, automation of visual inspection process can improve the efficiency of production lines by overcoming at least one of the aforesaid drawbacks.
Automated visual inspection of industrial web materials has extremely high requirements as compared to other inspection problems. Generally the width of industrial web is about 1.6-2.0 meters and requires an array of CMOS/CCD photosensors to inspect the web that is moving at the speed of 10-15 meters per minute. Consequently the throughput for 100% inspection is tremendous (10-15 MB image data per second when using line-scan cameras) and therefore most feasible solutions require additional DSP hardware components and reduction in computational complexity.
Prior researchers have attempted to create visual inspection systems using different approaches. Researchers have frequently used textile web samples to model the general problem of defect detection in various textured materials. Segmentation of fabric defects from the inspection images using the edge detection, mean and standard deviations of sub blocks, gray level co-occurrence matrix, and autocorrelation of images and, Karhunen-Loeve transform has been detailed in the literature. Solutions to defect detection problems using various texture analysis approaches include Gauss Markov Random Field (GMRF) modeling, Gabor filters, Wavelet transform, and Optical Fourier transform.
The periodic structure of woven web materials provides valuable information, and therefore Fourier domain features have been used in the fabric defect detection. This approach is extensively detailed in U.S. Pat. No. 4,124,300 issued on Nov. 7, 1976. When defects cause global distortions in woven web materials, Fourier analysis is most suitable. But this is not true for local fabric defects and therefore techniques that can simultaneously measure in spatial and spatial-frequency domain are more useful. U.S. Pat. No. 5,815,198 issued in December 1998 discloses a multi-scale wavelet decomposition of web images for fabric defect detection. The disadvantage with the techniques based on multiscale wavelet decomposition is that the wavelet bases (unlike RGF) are shift variant. Therefore, it is difficult to characterize a texture pattern from the wavelet coefficients since the wavelet descriptors depend on pattern location.
Additional examples include the U.S. Pat. No. 5,665,907 issued on Sep. 9, 1997 to the University of Chicago providing an ultrasonic system to detect fabric defects. Another recent U.S. Pat. No. 6,023,334 issued on Feb. 8, 2000 to Toshiba Engineering Corp. discloses an alternative approach based on brightness information, suited to inspection of homogenous surface like plain aluminum sheet or plain glass.
The aforementioned approaches suffer from various drawbacks such as the limited range of detectable defects, the extent of required computational and other resources as well as the need for further automation.
SUMMARY OF THE INVENTION
The present invention provides a new approach for defect detection using real Gabor function (RGF) filters that overcomes one or more of the drawbacks noted in the context of previous approaches. Thus, some defects that only alter the gray level arrangement of neighboring pixels (such as mispick in textile web) cannot be detected with the techniques based on mean and standard deviations of image sub-blocks. In light of the preceding, a multi-channel filtering approach to segment, i.e., identify, both fine and coarse texture defects is found to be useful as is shown later. This is achieved by segmenting the fine and coarse defects with the RGF filters of different scales (multichannel).
In an aspect of the invention, an embodiment of the invention uses real Gabor functions instead of complex Gabor function, thus, halving the computational time for the computation of the filtered images. Advantageously, the real Gabor function acts as a blob detector. In many embodiments of the invention, the real and imaginary parts of Gabor function are Hilbert pairs, thus, reducing the need for the imaginary part.
In another aspect, an embodiment of the invention enables automated selection of center frequency of real Gabor function using fast Fourier transform (FFT). Thus, change in the texture background under the inspection requires little or no manual tuning.
In another aspect, an embodiment of the invention exhibits superior computational and performance gain by the usage of non-linear function to estimate the local energy in the course of detecting defects.
Moreover, many embodiments of the invention employ the image fusion technique based on the Bemouli's rule of combination for integrating information from different channels to improve the detection rate while reducing false alarms.
In yet another aspect, the invention enhances computational efficiency by a judicious choice of thresholds and convolution masks to reflect appropriate trade-offs. Moreover, a simple thresholding method removes isolated noisy pixels without requiring any morphological operations.
REFERENCES:
patent: 4630304 (1986-12-01), Borth et al.
patent: 5301129 (1994-04-01), McKaughan et al.
patent: 5737072 (1998-04-01), Emery et al.
patent: 5740048 (1998-04-01), Abel et al.
patent: 5774177 (1998-06-01), Lane
patent: 5825501 (1998-10-01), Mee et al.
patent: 0 974 831 (2000-01-01), None
Partial European Search Report for EP 01 10 9375 dated Jun. 22, 2001.
Abdulghafour et al., “Non-deterministic Approaches in Date Fusion—A Review,”SPIE1383:596-610 (1990).
Beck et al., “Spatial Frequency Channels and Perceptual Grouping in Texture Segregation,”Comput. Vis. Graph. Image Process., 37: 299-325 (1987).
Bovik et al., “Multichannel Texture Analysis Using Localized Spatial Filters,”IEEE Trans. Pattern Anal. Mach. Intell., 12(1):55-73 (1990).
Casasent et al., “Detection Filters and Algorithm Fusion for ATR,”IEEE Trans. Image Process., 6(1):114-125 (1997).
Casasent, “Gabor Wavelet Filters and Fusion for Distortion-Invariant Multi-Class Object Detection,”Proc. SPIE, 2491:430-440 (1995).
Casasent, “Neural net design of macro Gabor wavelet filters for distortion-invariant object detection in clutter,”Opt. Eng., 33(7):2264-2271 (1994).
Casasent et al., “Real, imaginary, and clutter Gabor filter fusion for detection with reduced false alarms,”Opt. Eng., 33(7):2255-2363 (1994).
Chan et al., “Fabric Defect Detection by Fourier Analysis,”IEEE Trans. Ind. Appl., 36(5):1267-1276 (2000).
Cohen et al., “Automated Inspection of Textile Fabrics Using Textural Models,”IEEE Trans. Anal. Mach., 13(8):803-808 (1991).
Crounse et al., “Methods for Image Processing and Pattern Formation,”IEEE Trans. Circuits Syst., I-Fundam. Theory Appl., 42(10):583-601 (1995).
Escofet et al., “Detection of local defects in textile webs using Gabor filters,”Opt. Eng., 37(8):2297-2307 (1998).
Escofet et al., “Detection of Local Defects in Textile Webs Using Gabor Filters,”Proc. SPIE, 2785:163-170 (1996).
Jain et al., “Learning Texture Discrimination Masks,”IEEE Trans. Pattern anal. Mach. Intell., 18(2):195-205 (1996).
Jain et al., “Unsupervised Texture Segmentation Using Gabor Filters,”Pattern Recogn., 24(12):1167-1169 (1991).
Julesz, “Textons, the elements of texture perception, and their interactions,”Nature London, 290:91-97 (1981).
Namuduri et al., “Efficient Computation of Gabor Filter Based Multiresolution Response,”Pattern Recogn., 27
Kumar Ajay
Pang Kwok-Hung Grantham
Dickstein Shapiro Morin & Oshinsky LLP.
Font Frank G.
Punnoose Roy M.
The University of Hong Kong
LandOfFree
Defect detection system for quality assurance using... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Defect detection system for quality assurance using..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Defect detection system for quality assurance using... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3356520