METHOD AND APPARATUS FOR ANALYZING AN IMAGE TO DIRECT AND IDENTIFY PATTERNS

• US 20020054694A1
• Filed: 03/26/1999
• Published: 05/09/2002
• Est. Priority Date: 03/26/1999
• Status: Active Grant

• Current Assignee: Georgia Tech Research Corporation (University System of Georgia)
• Original Assignee: Georgia Tech Research Corporation (University System of Georgia)
• Inventors: WANG, PENG, ECHAUZ, JAVIER, MUFTI, MUID, VACHTSEVANOS, GEORGE J., DORRITY, J. LEWIS
• Granted Patent: US 6,650,779 B2
• Time in Patent Office: Days
• US Class Current: 382/111
• CPC Class Codes: G01N 21/898 Irregularities in textured or patterned surfaces, e.g. textiles, wood
• CPC Class Codes: G01N 21/956 Inspecting patterns on the surface of objects contactless testing of electronic circuits G01R31/308; testing currency G07D; manufacturing processes per se of semiconductor devices implementing a measuring step H01L22/10
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/30124 Fabrics; Textile; Paper
• CPC Class Codes: G06T 7/0006 using a design-rule based approach
• CPC Class Codes: G06V 10/421 by analysing segments intersecting the pattern
• CPC Class Codes: G06V 10/52 Scale-space analysis, e.g. wavelet analysis multi-scale boundary representations G06V10/42
• CPC Class Codes: G06V 2201/06 Recognition of objects for industrial automation

Abstract

A method and apparatus is provided which analyzes an image of an object to detect and identify defects in the object utilizing multi-dimensional wavelet neural networks.

Claims

1. An apparatus for analyzing a 2-D representation of an object, said apparatus comprising:

at least one sensor disposed to capture a 2-D representation of at least a portion of an object;

a memory that stores at least a portion of said 2-D representation; and

a processor that derives at least one signal from said 2-D representation, that generates a plurality of feature values representing features of said at least one signal and that provides said feature values to a multi-dimensional wavelet neural network which provides a classification output indicative of whether said representation comprises a predetermined pattern.

2. An apparatus according to claim 1 wherein said at least one sensor comprises a CCD array camera that captures said 2-D representation.

3. An apparatus according to claim 1 further comprising a light source that produces signals having a frequency in either the infared or visible range and wherein said at least one sensor comprises a vision camera responsive to said signals in the infared or visible frequency range.

4. An apparatus according to claim 3 further comprising a light diffuser disposed between said light source and said object for controlling the uniformity and intensity of the signals being projected onto the object.

5. An apparatus according to claim 1 comprising an analog to digital converter that converts said 2-D representation to a digital representation and provides said digital representation to said memory, and wherein said processor derives said at least one signal from said digital representation.

6. An apparatus according to claim 1 wherein said processor derives said at least one signal by projecting said 2-D representation into a 1-D signal.

7. An apparatus according to claim 6 wherein said processor derives said 1-D signal from said 2-D representation by utilizing a projection from the group consisting of vertical scanning, horizontal scanning and fractal scanning.

8. An apparatus according to claim 6 wherein said processor generates some of said feature values by deriving values from said 1-D signal.

9. An apparatus according to claim 6 wherein said processor generates some of said feature values by convolving said 1-D signal with a plurality of wavelet functions each employing a fast wavelet transform.

10. An apparatus according to claim 6 wherein said processor generates a feature vector by storing said feature values in an array.

11. An apparatus according to claim 1 wherein said processor determines whether at least one of said feature values is within a range indicative of a predetermined pattern and does not provide said feature values to said multi-dimensional wavelet neural network if said at least one feature value is not within said range.

12. An apparatus according to claim 1 wherein said multi-dimensional wavelet neural network comprises a plurality of wavelet neurons each defined by ψ

_u,b={square root}{square root over (diag(a)|)}ψ

(diag(a)(x−

b)) where x is a vector comprising said feature values, a is a squashing matrix for that neuron and b is the translation vector for that neuron.

13. An apparatus according to claim 1 further comprising a decision logic module that receives said classification output from said multi-dimensional wavelet neural network and said feature values to determine the degree of certainty that said pattern recognition system believes that said predetermined pattern was detected.

14. An apparatus according to claim 1 wherein the object is textile fabric being manufactured by warp knitting.

15. An apparatus according to claim 1 wherein the object is griege fabric.

16. An apparatus according to claim 1 wherein the object is textile fabric being inspected on a device selected from a group consisting of a circular loom, loom, an off-line inspection station or a weaving machine.

18. A method according to claim 17 wherein said predetermined pattern is one of a group of predetermined patterns and said multi-dimensional wavelet neural network comprises a plurality of wavelet neurons, wherein the step of providing from said multi-dimensional wavelet neural network comprises a step of determining what classification output to provide by providing an output of each of said neurons to a competition function, wherein said competition function provides said classification output.

19. A method according to claim 17 further comprising a step of deriving a 1-D signal from said 2-D representation and wherein the step of extracting a plurality of features from said 2-D digital representation comprises extracting a plurality of feature values from said 1-D signal.

20. A method according to claim 19 wherein said step of extracting said plurality of feature values comprises deriving at least one feature value of said plurality of feature values from said 1-D signal.

21. A method according to claim 19 wherein said step of extracting said plurality of feature values comprises a step convolving said 1-D signal with a plurality of wavelet functions utilizing fast wavelet transforms to produce a plurality of fast wavelet coefficients corresponding to at least one feature value of said plurality of features.

22. A method according to claim 19 wherein said step of extracting said plurality of feature values comprises the steps of:

deriving at least one feature value of said plurality of feature values from said 1-D signal,

convolving said 1-D signal with a plurality of wavelet functions utilizing fast wavelet transforms to produce a plurality of wavelet coefficients corresponding to at least one feature value of said plurality of features, and

arranging said plurality of feature values into a feature vector.

23. The method of claim 17 further comprising a step of determining whether at least one of said plurality of features comprise a value indicative of a predetermined pattern, and if said value is not indicative of a predetermined pattern not providing said feature values to said multi-dimensional wavelet neural network.

24. The method of claim 17 wherein said object comprises a textile material.

26. The computer readable medium of claim 25 further comprising means for instructing the computer to generate a 1-D representation of said 2-D digital image, wherein said feature vector is generated from said 1-D representation.

27. The computer readable medium of claim 25 wherein said means for instructing the computer to generate a feature vector comprises means for instructing the computer to derive said feature values from said 1-D signal.

28. A computer readable medium of claim 25 wherein said means for instructing the computer to generate a feature vector comprises means for instructing the computer to convolve said 1-D signal with a plurality of fast wavelet functions.

29. A computer readable medium according to claim 25 further comprising means for instructing said computer to generate a feature library comprising a plurality of selected features, wherein said computer is instructed by said means for instructing the computer to generate said feature vector so that the components of said feature vector each correspond to a value of one of said selected features.

30. A computer readable medium according to claim 25 wherein said multi-dimensional wavelet neural network comprises a plurality of wavelet functions, the computer readable medium further comprising means for instructing said computer to adjust said wavelet functions based upon known feature vectors of said predetermined patterns.

31. A computer readable medium according to claim 25 further comprising means for instructing the computer to determine whether said feature vector comprise at least one value indicative of a predetermined pattern, and if said at least one value is not indicative of a predetermined pattern instructing said computer not provide said feature vector to said multi-dimensional wavelet neural network.

17. A method for pattern recognition, comprising:

generating a 2-D digital representation of at least part of an object, the digital representation comprising a plurality of pixels;

extracting a plurality of feature values from said 2-D digital representation of said at least part of said object;

providing said plurality of feature values to a multi-dimensional wavelet neural network; and

providing from said multi-dimensional wavelet neural network a classification output indicative of a predetermined pattern if said feature values are indicative of a predetermined pattern.

25. A computer readable medium containing instructions for a computer comprising:

means for instructing the computer to read at least a portion of a 2-D digital image, said digital image comprising a plurality of pixels;

means for instructing the computer to generate a feature vector from said digital image;

means for instructing the computer to provide said feature vector to a multi-dimensional wavelet neural network; and

means for instructing the computer to provide a classification output indicative of a predetermined pattern from said multi-dimensional neural network if said feature values are indicative of a predetermined pattern.

32. An apparatus for pattern recognition comprising:

an input that receives a 2-D representation of at least part of an object;

a memory that stores at least a portion of said 2-D representation; and

a processor that generates a plurality of feature values representing features of said at least one signal and that provides each of said feature values to a perceptron neural network comprising a plurality of neurons each defined by the function ψ

_a,b={square root}{square root over (|diag(a)|)}ψ

(diag(a)(x−

b)) where x is a vector comprising said feature values, a is a squashing matrix for the neuron and b is the translation vector for that neuron, said perceptron neural network providing a classification output indicative of whether said representation contains a predetermined pattern.

33. An apparatus according to claim 32 wherein said processor derives at least one 1-D signal from said 2-D representation and said processor derives said feature values from said at least one 1-D signal.

34. An apparatus according to claim 33 wherein said processor derives said 1-D signal from said 2-D representation by utilizing a projection from the group consisting of vertical scanning, horizontal scanning and fractal scanning.

35. An apparatus according to claim 33 wherein said processor generates some of said feature values by deriving values from said 1-D signal.

36. An apparatus according to claim 33 wherein said processor generates some of said feature values by convolving said 1-D signal with a plurality of fast wavelet functions.

37. An apparatus according to claim 33 wherein said processor generates a feature vector by storing said feature values in an array.

38. An apparatus according to claim 32 wherein said processor determines whether at least one of said feature values is within a range indicative of a predetermined pattern and does not provide said feature values to said multi-dimensional wavelet neural network if said at least one feature value is not within said range.

39. An apparatus according to claim 32 further comprising decision logic module that receives said classification signal from said perceptron neural network to determine the degree of certainty of the detection of said predetermined pattern.

40. An apparatus according to claim 32 wherein said object comprises a textile fabric.