Method and apparatus for analyzing an image to detect and identify defects

• US 5,815,198 A
• Filed: 05/31/1996
• Issued: 09/29/1998
• Est. Priority Date: 05/31/1996
• Status: Expired due to Term

• Current Assignee: Georgia Tech Research Corporation (University System of Georgia)
• Original Assignee: Dorrity J. Lewis, George J. Vachtsevanos, Muid Mufti
• Inventors: Mufti, Muid, Vachtsevanos, George J., Dorrity, J. Lewis
• Primary Examiner(s): Chin, Tommy P.
• Assistant Examiner(s): Le, Vu
• Application Number:
  US08/657,023
• Time in Patent Office: 851 Days
• Field of Search: 348/86, 348/87, 348/88, 348/91, 348/92, 348/125, 348/126, 348/128, 348/129, 348/130, 348/131, 382/118, 382/156, 382/190, 382/149, 382/141
• US Class Current: 348/88
• CPC Class Codes: G01N 21/88 Investigating the presence of flaws or contamination
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 2207/20068 Projection on vertical or horizontal image axis
• CPC Class Codes: G06T 2207/30124 Fabrics; Textile; Paper
• CPC Class Codes: G06T 7/001 using an image reference approach

Abstract

The Invention includes a method and apparatus which analyzes an image of an object to detect and identify defects in the object. The present invention utilizes a scanning technique which converts a 2-D image of the object into a 1-D image, a transformation technique which extracts relevant features from the image, and a fuzzy inferencing technique which utilizes the features generated by the transformation technique to detect and identify defects. The present invention preferably includes an off-line learning process which selects the optimum transform coefficients for a given set of defects and stores the corresponding features in a rulebase. Preferably, the wavelet transform is used as the transformation technique to provide an analysis of the image which is localized in the frequency and time domains. The present invention may also include an on-line learning process when the present invention is incorporated into a manufacturing process for real time inspection of the object being manufactured. The on-line learning process attempts to nullify the effects of noise which may be associated with the image sensors being used to read the image by using a similarity function to maintain a predetermined level of fuzziness of the inference engine of the present invention.

Claims

1. An apparatus for analyzing an image to detect and identify defects in an object, said apparatus comprising:

at least one image sensor disposed to capture an image of the object;

a light source for projecting light onto the object;

means for converting the captured image into a digital representation;

a storage device for receiving the digital representation of the image from said means for converting the captured image and for storing the digital representation; and

processing means for processing the digital representation of the captured image stored in said storage device, wherein the processing comprises a convolution operation on the digital representation whereby one or more wavelets are used to transform the input digital signal into wavelet coefficients thus extracting certain defect feature sets from the digital representation, performing fuzzification by calculating the degree of membership of the extracted feature sets in feature templates stored in a knowledge base in said storage device in accordance with a similarity function, generating fuzzy sets from said defect feature sets if their degree of membership exceeds a threshold, and performing fuzzy inferencing with said rules and fuzzy sets to classify the type of defect that has occurred.

2. An apparatus according to claim 1 wherein prior to performing the transformation, a horizontal, a vertical and a fractal scanning technique are performed on the digital representation to convert the digital representation into several 1-D digital representations.

3. An apparatus according to claim 2 wherein said at least one image sensor is comprised of two or more CCD array cameras, and wherein each CCD array camera captures an image of a different area of the object and wherein a defect is determined to exist when a defect is detected in corresponding scan lines of at least two of said CCD array cameras.

4. An apparatus according to claim 3 wherein the object is a fabric comprising one of a griege fabric, a textile fabric being manufactured on a loom, a textile fabric being manufactured by warp knitting, and a textile fabric being manufactured on a circular loom.

5. An apparatus according to claim 2 wherein said at least one image sensor is comprised of a line scan camera and wherein a full image of the object is obtained by producing relative motion between said line scan camera and the object and by accumulating line scans to obtain a full image of the object and wherein the full image is then analyzed to determine whether a defect exists and, if so, the type of defect.

6. An apparatus according to claim 5 wherein the object is a fabric comprising one of a griege fabric a textile fabric, being manufactured on a loom, a textile fabric being manufactured by warp knitting and a textile fabric being manufactured on a circular loom.

7. An apparatus according to claim 1 wherein said at least one image sensor is comprised of two or more CCD array cameras, and wherein each CCD array camera captures an image of a different area of the object and wherein a defect is determined to exist when a defect is detected in corresponding scan lines of at least two of said CCD array cameras.

8. An apparatus according to claim 1 wherein said at least one image sensor is comprised of a line scan camera and wherein a full image of the object is obtained by producing relative motion between said line scan camera and the object and by accumulating line scans to obtain a full image of the object and wherein the full image is then analyzed to determine whether a defect exists and, if so, the type of defect.

9. An apparatus according to claim 1 wherein the object is a fabric comprising one of a griege fabric, a textile fabric being manufactured on a loom, a textile fabric being manufactured by warp knitting, and a textile fabric being manufactured on a circular loom.

10. An apparatus according to claim 1 wherein said light source produces only infrared light and wherein said at least one image sensor is an infrared camera.

11. An apparatus according to claim 1 further comprising a light diffuser disposed between said light source and the object for controlling the uniformity and intensity of the light being projected onto the object.

12. A method for detecting and identifying defects in an object comprising the steps of:

storing the digital representation of the image in memory;

transforming the digital representation by performing a transformation process which extracts certain defect feature sets contained in the image of the object;

the transforming step including convolving the digital representation with a series of wavelets;

wherein the resulting wavelet coefficients comprise the defect features;

creating fuzzy sets by calculating the degree of membership of the extracted features with stored feature templates according to a similarity function; and

using the fuzzy sets in a fuzzy inferencing scheme to infer and classify whether a defect has been detected and, if so, the type of defect that has occurred;

wherein the fuzzy wavelet coefficients are compared with fuzzy sets stored as representative features in a knowledge base for different wavelet scales and for a plurality of anticipated defects and wherein the wavelet scales are selected to maximize detectability and identifiability measures.

13. The method of claim 12 wherein the step of selecting the wavelet scales is accomplished by using a genetic algorithm.

14. The method of claim 12 further comprising an on-line learning process which is accomplished by obtaining statistical information relating to said image and by scaling said image by a factor determined in accordance with the statistical information.

15. A method according to claim 12 wherein a horizontal, a vertical and a fractal scanning technique are performed on the digital representation stored in memory to convert the digital representation into several 1-D digital representations.

16. An apparatus for analyzing an image of a fabric to detect and identify defects in the fabric, said apparatus comprising:

at least one image sensor disposed to capture the image of at least part of the fabric;

a light source for projecting light through the fabric;

means for converting the captured image into a 2D digital representation;

a storage device for receiving said 2D digital representation of the image from said converting means and for storing said digital representation; and

processing means for processing said 2D digital representation of the captured image, said processing means including means for converting said 2D digital representation into at least one ID scan of the image which is particularly adapted to detect an expected defect type, means for convolving said at least one 1D scan of said image with at least one predetermined wavelet from a firmly of wavelets to extract certain defect features in the form of wavelet coefficients from said at east one 1D digital representation means for concerting said wavelet coefficients into fuzzy sets via a similarity transformation, and means for declaring and classifying defects based upon said defect features.

17. An apparatus for analyzing an image as set forth in claim 16, wherein said apparatus further comprises;

means for preprocessing said image to improve the defect signature contrast with respect to background noise.

18. An apparatus for analyzing an image as set forth in claim 16, wherein said apparatus further comprises:

means for on-line learning which adjusts the defect feature detection parameters of the apparatus based upon measured variables of the process in real time.

19. An apparatus for analyzing an image as set forth in claim 18, wherein said on-line learning means further includes:

means for modifying the wavelet parameters.

20. An apparatus for analyzing an image as set forth in claim 18, wherein said on-line learning means further includes:

means for modifying the fuzzy membership function parameters.

21. An apparatus for analyzing an image as set forth in claim 16, wherein said apparatus further comprises:

means for off-line learning which adjusts one of the detection and classification parameters of the apparatus based upon the measured variables of the process.

22. An apparatus for analyzing an image as set forth in claim 21, wherein said off-line learning means further includes:

means for modifying the fuzzy defect rules.

23. An apparatus for analyzing an image as set forth in claim 16, wherein said apparatus further includes:

means for estimating the noise in the image and for adjusting one of the detection and classification parameters of the apparatus based on the measured noise.

24. An apparatus for analyzing an image as set forth in claim 16, wherein said means for converting said 2D digital representation further includes:

means for horizontally scanning said 2D digital representation which is adapted for sensitivity to fabric warp defects;

means for vertically scanning said 2D digital representation which is adapted for sensitivity to fabric filling defects; and

means for fractal scanning said 2D digital representation which is adapted for sensitivity to fabric area defects.