Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
2001-05-17
2004-09-28
Couso, Yon J. (Department: 2623)
Image analysis
Pattern recognition
Feature extraction
C382S266000, C382S264000
Reexamination Certificate
active
06798910
ABSTRACT:
FIELD OF THE INVENTION
The invention is generally related to computers and computer software. More specifically, the invention is generally related to a manner of edge detection in image processing.
BACKGROUND OF THE INVENTION
Traditional edge detection methods have generally required a high enough signal to noise ratio and fine enough image clarity that the transition from one region to another does not significantly exceed a single pixel in width. Even edge detection methods that have been adapted to noisy images still appear to carry the assumption that the noise is superimposed on top of an image with edges that do not exceed a single pixel in width.
In essence, edge detection has traditionally been a form of high-pass filtering, although in noisy images it performs better when implemented as a band-pass filter process. Since speckle noise has a definite high frequency component, the best results will naturally come from a process that excludes as much noise energy from the detection process as possible. In certain images, such as highly magnified images, low light images, or pictures taken of a moving object or where the camera is moved, the displayed resolution creates a non-negligible spatial auto-correlation among the pixel amplitudes. Thus, edges become border regions with non-zero width.
Images that have blur or noise conventionally cannot receive the benefits of edge detection. For example, digital photography effects such as embossing or conversion to a line drawing are not readily available. Certain artificial intelligence applications rely upon interpreting a scene. In some application, these limitations are partially offset by having knowledge of the expected shape and edge thickness of objects within a digital image so that a tailor-made template may be used for detection.
Consequently, a significant need exists for a way to tune the spectral response of edge detection to accommodate only the bandwidth of a natural edge in a particular image, and reject as much of the high frequency noise energy as possible, yet not require beforehand knowledge of the characteristics of the image clarity.
SUMMARY OF THE INVENTION
The invention addresses these and other problems associated with the prior art by providing an apparatus, program product and method in which “Wilson” horizontal and vertical edge detection kernels are formed of various sizes, each size sensitive to detecting edge widths of various sizes. Repeating convolution of a digital image with various sizes of Wilson kernels achieves edge detection even when the size of the edge is not known in advance.
In one aspect of the invention, an image is subjected to a series of edge detection processes using kernels tailored to border regions of increasing width, until the natural edge width is found. In blurred images, such a process yields improving results with successive iterations until the natural edge width for that particular image is reached. Further increasing the width of the kernel does not yield significant improvement, but rather begins to cause a loss of features, so the process is then halted. There are at least two advantages to this approach. First, since the method is noise tolerant, edges may be found in images otherwise too noisy or coarse for traditional approaches. Second, a measure of the natural edge width quantifies the blur in the image and acts as a metric for clarity.
These and other advantages and features, which characterize the invention, are set forth in the claims annexed hereto and forming a further part hereof. However, for a better understanding of the invention, and of the advantages and objectives attained through its use, reference should be made to the Drawings, and to the accompanying descriptive matter, in which there is described exemplary embodiments of the invention.
REFERENCES:
patent: 4330833 (1982-05-01), Pratt et al.
patent: 4720871 (1988-01-01), Chambers
patent: 4918742 (1990-04-01), Simonds
patent: 5022091 (1991-06-01), Carlson
patent: 5081692 (1992-01-01), Kwon et al.
patent: 5748792 (1998-05-01), Wober
patent: 6064768 (2000-05-01), Hajj et al.
patent: 6072904 (2000-06-01), Desai et al.
patent: 6094511 (2000-07-01), Metcalfe et al.
patent: 6687300 (2004-02-01), Fujita et al.
Fong-chao Wu, Comparison of Edge Detectors, 1-12, www-aaa.eng.ohio-state.edu/~wuf/study/paper/content.htm (Sep. 12, 2000).
Other Methods of Edge Detection, 1-7, www.owlnet.rice.edu/~elec539/projects97/morphjrks/moredge.html (Sep. 12, 2000).
Jordan, Ramiro et al., Edge Detection II, 1-3, sevilleta.unm.edu/~bmilne/khoros/html-dip/c9/s2/front-page.html (Sep. 12, 2000).
Image Processing Khorosware; Edge Detection I, 1-6, www.ee.ghu.ac.il/~greg/graphics/special.html (Sep. 12, 2000).
Couso Yon J.
Franklin David E.
Hollins Gerald B.
Kundert Thomas L.
The United States of America as represented by the Secretary of
LandOfFree
Self-optimizing edge detection in blurred, high-noise images does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Self-optimizing edge detection in blurred, high-noise images, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Self-optimizing edge detection in blurred, high-noise images will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3209164