Image analysis – Image enhancement or restoration – Artifact removal or suppression
Reexamination Certificate
1998-09-18
2001-05-15
Lee, Thomas D. (Department: 2624)
Image analysis
Image enhancement or restoration
Artifact removal or suppression
C382S254000, C382S264000, C382S205000
Reexamination Certificate
active
06233364
ABSTRACT:
COPYRIGHT NOTICE
A portion of the disclosure recited in this specification contains material which is subject to copyright protection. Specifically, a Microfiche Appendix in accordance with 37 CFR Section 1.96 is included that lists source code instructions for a process by which the present invention is practiced in a computer system. The Microfiche Appendix comprises two sheets of microfiche containing 166 frames, or pages, of source code. The copyright owner has no objection to the facsimile reproduction of the specification as filed in the Patent and Trademark Office. Otherwise all copyright rights are reserved.
BACKGROUND OF THE INVENTION
This invention relates in general to computer image processing systems and more specifically to a system to identify and tag portions of a digital image representing dust or scratches on the original hard-copy image, in order that these images of dust may then be erased from the digital image.
It is a well known problem in image processing that particles of dust may accumulate on a hard-copy positive or negative version of an image, or that the hard copy may become scratched before a secondary copy of the image is made by a camera or a scanner. Conversion of a photographic negative or positive image to a digital image suitable for processing on a computer requires such a secondary copy to be made of the image, and hence it is common for digital images to be scratched or polluted by dust.
The dust pollution problem may be alleviated either by early-detection or late-detection techniques. Early-detection techniques, such as are taught in U.S. Pat. No. 4,189,235, entitled “Test Device for Dynamically Measuring the Degree of Dirt Accumulation on Bank-Notes” by Guter et al., detect dust on an original hard copy image before it is scanned, thus permitting the image to be cleaned by mechanical means before scanning. In contrast, late-detection techniques detect the representations in the digitized image of dust on the original image with the intent of then letting the user of the computer system retouch the image so as to remove the dust. Typically, early-detection techniques require special hardware, not part of a normal digital image-processing setup. In contrast, late-detection techniques use only the standard digital image-processing hardware. Late-detection techniques use the main computer of the image-processing setup to identify representations of dust in the digital image, thus permitting the dust to be removed from the image either by manual intervention by the computer operator or by additional computer software. The dust-detection techniques of this invention are late-detection techniques.
A patent which teaches a technique which is a hybrid between early-detection techniques and late detection techniques is U.S. Pat. No. 5,436,979 entitled “Process for Detecting and Mapping Dirt on the Surface of a Photographic Element” by Gray et al. This patent describes a technique which is like the late-detection techniques in that it uses conventional image-processing hardware and software to scan an image and to detect digitized representation of dust in the digital image. However, the invention scans a blank image (a blank piece of film) with the intent that if dust is detected on the blank image the operator of the image scanner may adjust the scanner's cleaning process to minimize the level of dust on subsequent non-blank images. The “blank” images scanned by the cited invention are not featureless. They may be, for example, pieces of unexposed film containing the graininess of the film itself. Hence the problem of identifying the scanned image of dust against the scanned grainy background has something in common with the problem of identifying the scanned image of dust against a scanned developed piece of film. The patent describes a process of (a) first blurring the digital image, and then (b) taking the difference between the blurred image and the original image, thus forming a residual difference image in which anything in the residual image that is sufficiently different from the blurred image can be presumed to be an anomaly. Any such anomaly is then a candidate to be deemed a piece of dust (after further testing).
As described in U.S. Pat. No. 5,436,979 cited above, the process of looking for dust in a digitized image is the process of looking for anomalies in the image. The same is true of the process of looking for scratches. Hence practitioners of the art of dust or scratch detection in images can learn something from other arts that detect anomalies in images such as, in particular, medical imaging software for detecting images of tumors. A patent teaching the latter art is U.S. Pat. No. 4,907,156, entitled “Method and System for Enhancement and Detection of Abnormal Anatomic Regions in a Digital Image” by Doi et al.
U.S. Pat. No. 4,907,156 on “Abnormal Anatomic Regions” is like U.S. Pat. No. 5,436,979 on “Detecting and Mapping Dust” in that it finds candidate anomalies by forming the difference between an original image (or a simple derivative thereof) and a blurred version of the original image. Pixels (picture elements) with extreme values in the difference image represent pixels whose values differ a lot from the values of their neighbors in the original image and hence are candidates to be anomalies. The specific technique in the “Anatomic Regions” patent forms an SNR-maximized image (where “SNR” denotes the signal to noise ratio), in which anomalous features such as tumors will be emphasized, and also a blurred or SNR-minimized image. A difference image is formed, representing the difference between the SNR-maximized image and the SNR-minimized image. Pixels or groups of pixels with extreme values in the difference image are candidates to be deemed images of tumors. Subsequent additional image processing, using “circularity, size and growth tests”, is then used to further characterize the already-detected candidate tumors.
Difference techniques which compare pixel values to the mean of pixel values in the local neighborhood, such as described above, tend to do a good job of detecting anomalies (dust, scratches or tumors) against a uniform or smoothly varying background, such as sky, but a poor job of detecting anomalies against a varied or chaotic background, such as the leaves of a tree in sunlight, or a pebbly beach. In the latter case, the difference techniques described above will have a great number of false positives, i.e., groups of pixels falsely deemed to be anomalies. This is because it is characteristic of a chaotic portion of an image that it has lots of pixels whose values differ greatly from the neighborhood mean value. This characteristic is precisely the distinguishing characteristic of anomalies under the standard difference techniques. Hence these techniques tend to produce an inordinately large number of false positives in chaotic regions of the image. A result is that when using such techniques it is necessary to do a lot of post-processing, either manually by looking at an image, or with additional computer processing, to discard as dust or scratch candidates these false positive pixels or groups of pixels.
SUMMARY OF THE INVENTION
The present invention provides a system for detecting anomalies such as dust or scratches on an image, such as a photograph, and overcomes some of the problems of the prior art approaches. For example, the approach of the present invention results in fewer false detections. The principal idea underlying the present invention's technique for reducing the number of false positives is that it should take a much stronger indicator to cause a pixel to be tagged as a dust or scratch candidate in a chaotic region of the image (e.g., a pebbly beach), than in a smooth or smoothly varying region of the image (e.g., the sky). The present invention achieves the goal of reducing the number of first-pass false positives by taking as its first candidates for dust or scratch particles not those pixel regions whose values stand out from the mean of the values of neighboring pixels,
Krainiouk Alexei
Minner Richard T.
Dainippon Screen Engineering of America Incorporated
Kulas Charles J.
Lee Thomas D.
Townsend and Townsend / and Crew LLP
LandOfFree
Method and system for detecting and tagging dust and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and system for detecting and tagging dust and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for detecting and tagging dust and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2512393