Image analysis – Image compression or coding – Pyramid – hierarchy – or tree structure
Reexamination Certificate
2000-06-08
2004-03-09
Couso, Jose L. (Department: 2721)
Image analysis
Image compression or coding
Pyramid, hierarchy, or tree structure
Reexamination Certificate
active
06704454
ABSTRACT:
The invention relates generally to an apparatus and a concomitant method for image processing and, more particularly, to an apparatus and method using a model for computing probability distributions of images, which, in turn, can be applied to image processing applications, such as object recognition, object classification and the like.
BACKGROUND OF THE DISCLOSURE
Current approaches to object recognition estimate the probability of a particular class given an image, Pr(class|image), i.e., the probability that, given an image, it is an image of an object of a particular class. For example, in mammography, given an image, the Pr(class|image) can be a probability of a class, i.e., the class can be a “tumor” or “non-tumor.” However, such an approach is suspect to erroneous classification of an image. Additionally, this approach will likely fail to account for the detection and rejection of unusual images.
To account for unusual images and reduce erroneous classification of images, the object recognition approaches require a better model for an image probability distribution or a probability distribution of images. Given this image probability distribution, it is possible to provide enhanced object recognition, by training a distribution for each object class and using Baye's Rule of conditional probability to obtain Pr(class|image), where Pr(class|image)=Pr(image|class)Pr(class)/Pr(image).
Current image distribution methods have produced positive results for textures, but fail to adequately capture the appearance of more structured objects in the image. Namely, these methods merely capture local dependencies or correlations in images, but fail to capture non-local and long-range dependencies. As such, these methods fail to adequately represent the image probability distribution or the probability distribution of images.
Therefore, a need exists in the art for an apparatus and a concomitant method that provides an image probability distribution that captures non-local and long-range dependencies of an image. Such an image probability distribution would enhance a variety of image processing applications. For example, the image probability distribution would enable the detection and rejection of unusual images in object recognition systems.
SUMMARY OF THE INVENTION
The present invention is an apparatus and method to compute an image probability distribution or a probability distribution of images. Namely, the present invention performs Hierarchical Image Probability (HIP) modeling to provide the image probability distribution.
More specifically, the present invention decomposes the input image into a low-pass, i.e., gaussian, pyramid from which one or more feature pyramids and subsampled feature pyramids are derived. These pyramids form a hierarchical representation that models local information of the image. Next, the non-local information in the image are modeled with a plurality of labels or hidden variables. The plurality of labels and at least one of the feature and subsampled feature pyramids are used to compute the image probability distribution. In one embodiment of the invention, the image probability distribution is used in an object detection system.
REFERENCES:
patent: 4751742 (1988-06-01), Meeker
patent: 5179441 (1993-01-01), Anderson et al.
patent: 5321776 (1994-06-01), Shapiro
patent: 5661822 (1997-08-01), Knowles et al.
patent: 5671294 (1997-09-01), Rogers et al.
patent: 5799100 (1998-08-01), Clarke et al.
patent: 5835034 (1998-11-01), Seroussi et al.
patent: 5995668 (1999-11-01), Corset et al.
patent: 6018728 (2000-01-01), Spence et al.
patent: 6049630 (2000-04-01), Wang et al.
patent: 6141446 (2000-10-01), Boliek et al.
patent: 6263103 (2001-07-01), Freeman et al.
patent: 6324532 (2001-11-01), Spence et al.
patent: 6380934 (2002-04-01), Freeman et al.
patent: 6501861 (2002-12-01), Cho et al.
patent: 6549666 (2003-04-01), Schwartz
Crouse et al. “Wavelet-Based Statistical Signal Processing Using Hidden Markov Models” IEEE Transactions on Signal Processing, vol. 46, No. 4, pp. 886-902. Apr. 1998.
Parra Lucas
Spence Clay Douglas
Burke William J.
Couso Jose L.
Sarnoff Corporation
LandOfFree
Method and apparatus for image processing by generating... 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 apparatus for image processing by generating..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for image processing by generating... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3283389