Image analysis – Learning systems
Reexamination Certificate
2000-10-31
2004-01-13
Ahmed, Samir (Department: 2623)
Image analysis
Learning systems
C382S143000, C382S145000
Reexamination Certificate
active
06678404
ABSTRACT:
TECHNICAL FIELD
This invention relates to object detection and location using automatic referencing in a computer vision system.
BACKGROUND OF THE INVENTION
Many computer vision applications allow the user to specify an object of interest template (i.e. a type of reference image) and use the template to match new images for object of interest detection and/or location. This simple approach is the basis for many useful machine vision tools in object finding, location, alignment and measurements. To detect objects of interest, a correlation method is applied to compare the grayscale values of the images and the object template (Reference 4, Computer Vision p.66-68). The position having the highest correlation value is the detected object location. This approach works well when the objects of interest exhibit little difference from the template. However, it cannot accurately locate objects that change in orientation or scale from the template. It also performs poorly when subjected to illumination changes, production variations, and partial occlusion.
To circumvent these limitations, prior art uses normalization methods such as normalized grayscale correlation (Reference 4, Computer Vision p.66-68) to increase the robustness of the template matching approach in the presence of large background variations. However, these methods yield many false alarms. To overcome this difficulty, a geometric information approach is used in place of grayscale correlation (http://www.cognex.com/marketing/products/prod
—
8000_patmax.asp reference: PatMax). Examples of the geometric approach would be to encode a square as four line segments and a football as two arcs and measures characteristics such as shape, dimensions, angle, arcs, and shading. The geometric information from both the features and spatial relationships are used to detect objects of interest without regard to the object's orientation, size or appearance. Unfortunately, this approach depends on reliable identification and isolation of key geometric features within an image. It fails on objects with ambiguous or difficult to detect geometric features. Moreover, many false alarms occur when detection sensitivity is set to detect ambiguous objects at reasonable capture rates. Again, the difficulty is with objects having substantial differences from the template such as a large orientation difference from the template or having partial occlusion, scale changes, etc. Another difficulty is the problem of automating this approach, since prior art methods frequently rely on a trial and error manual programming. Finally, in the prior art it is difficult to know during the design process, all the variables that the system will actually encounter during use and assure that the full application range has been represented.
OBJECTS AND ADVANTAGES
This invention seeks to improve the conventional object matching approach such as the correlation method or geometric information method already known to those skilled in the art. Instead of relying on a single or a few templates specified by users, it is an object of this invention to automatically extract useful information from real application images that represent the objects of interest as well as their expected variations in the application specific domain. This systematic automatic approach produces data that better represents the application.
It is an object of this invention to improve efficiency and effectiveness by changing methods used to represent templates. The invention enhances detection and location efficiency and signal to noise ratio by representing templates as multi-resolution image data. Multi-resolution image representation facilitates a coarse to fine template matching approach and distributes image variations into different spatial-frequency bands, allowing specific weighting and discrimination filtering. Systematic adjustment of system parameters such as orientation, gain and offset, and noise level for effective matching can be done more rapidly or more effectively with multi-resolution image data.
SUMMARY OF THE INVENTION
In an embodiment, the invention includes an automatic reference based defect detection system having an automatic reference image generation module. The embodiment further comprises means for generating a mean reference image and a deviation reference image. Means are further provided for thresholding and normalizing a discrepancy image.
In an embodiment, the invention includes an automatic reference based object location system.
In an embodiment the invention includes an automatic reference image generation system.
In an embodiment, the reference images include a mean image and a deviation image. The mean image is an automatically derived representation of the template image under a specified “ideal” situation. The deviation image is an automatically derived representation of the expected variations on different portions of any template image. In an embodiment, the invention heightens the weights of object regions with low variations and lowers the weights of object regions with high variations in an object detection process. Using this weighting, the reference mean image enhances the detection signal of the object of interest and the reference deviation image decreases the effects of noise.
In an embodiment, the reference images are represented in a multi-resolution image space.
REFERENCES:
patent: 5046111 (1991-09-01), Cox et al.
patent: 5229868 (1993-07-01), Kanno et al.
patent: 5842194 (1998-11-01), Arbuckle
patent: 5850466 (1998-12-01), Schott
C.H. Anderson, C.R. Carlson, R.W. Kolpfenstein, “Spatial-frequency Representations of Images with Scale Invariant Properties”, SPIE vol 360:90-95.
Peter J. Burt, “Fast Algorithms for Estimating Local Image Properties”, Computer Vision, Graphics and Image Processing, vol. 21: 368-382, 1983.
E.H. Adelson, C.H. Anderson, J.R. Bergen, P.J. Burt, J.M. Ogden, “Pyramid Methods in Image Processing”, RCA Engineer 29-6, Nov./Dec. 1984.
Ballard DH and Brown CM, “Normalized Correlation”, Computer Vision, Prentice-Hall Inc., 1982.
Burt, PJ, “Fast Filter Transforms for Image Processing”, Comp. Graphics and Image Processing, 16:20-51, 1981.
Burt PJ and Adelson E, “The Laplacian Pyramid as a Compact Image Code” IEEE Trans. on Communication, vol. 31: 532-540, 1983.
Lee JSJ, Haralick RM and Shapiro LG, “Morphologic Edge Detection”, IEEE Trans. on Communication, Vol; 31: 532-540, 1983.
Maragos P, “Pattern Spectrum and Multiscale Shape Representation”, IEEE Trans on Pattern Analysis and Machine Intelligence, vol. 11, 7-12:701-716, 1989.
P.J. Burt, “The Pyramid as a Structure for Efficient Computation”, New York: Springer-Verlag, 1984.
James L. Crowley and Arthur C. Sanderson, “Multiple Resolution Representation and Probabilistic Matching of 2D Gray Scale Shape”, IEEE:95-105, 1984.
Sternberg SR, “Grayscale Morphology”, Computer Vision, Graphics, and Image Processing, Vol 35:333-355, 1986.
Lee Shih-Jong J.
Oh Seho
Ahmed Samir
Bali Vikkram
LandOfFree
Automatic referencing for computer vision applications does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Automatic referencing for computer vision applications, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Automatic referencing for computer vision applications will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3237766