Image analysis – Color image processing – Pattern recognition or classification using color
Reexamination Certificate
1999-04-20
2002-11-05
Mancuso, Joseph (Department: 2623)
Image analysis
Color image processing
Pattern recognition or classification using color
C382S203000
Reexamination Certificate
active
06477269
ABSTRACT:
FIELD OF THE INVENTION
This invention relates generally to computer images, and more specifically to a tool for assisting a user in finding images similar to a selected image.
BACKGROUND OF THE INVENTION
With today's technology, computer users have easy access to thousands of digital images. As technology continues to advance, more and more computer users will have access to more and more images. However, as the number of images to which computer users have access increases, so does the difficulty in locating a particular image.
Attempts have been made to solve this problem. One solution is the use of keywords. There are several difficulties with the use of keywords in trying to locate an image. First, all images must be manually tagged. Second, some visual aspects of an image are hard to describe. Third, users must guess which keyword(s) were used to tag images of potential interest.
Based on the problems with keyword searching, other approaches have been used to describe images so that a database of images can be searched. One approach is to obtain a wavelet signature based on a wavelet decomposition to query by example. A wavelet signature describes the shape of objects located in an image First, the wavelet signature of a selected image that contains one or more features of interest to the user (such as an image of an object) is determined. Then, wavelet signatures are determined for all potentially similar images that may be of interest to the user contained in an image catalog (i.e., an image database) to be searched. Next, the wavelet signatures of the potentially similar images are compared with the wavelet signature of the selected image. The closer the wavelet signature of a potentially similar image is to the wavelet signature of the selected image, the greater the likelihood that the images are similar. Potentially similar images are ranked using this criteria and the result displayed, in order, to the user. There are some problems with this approach to image searching, for example, if an image is rotated, the original image and the rotated image will produce significantly different wavelet signatures.
Another approach to image searching employs color histograms. A color histogram provides general information about the color characteristics of an image. In this approach, the color histogram of a selected image that contains one or more features of interest to the user is determined. Then color histograms of all potentially similar images contained in an image catalog to be searched that may be of interest to the user are determined. The color histograms are compared, ranked and displayed, in order, to the user. There are also problems with this approach, for example, a color, a black and white, and a gray scale version of the same image will produce significantly different color histograms. Accordingly, there exists a need for an easy method to more accurately locate images that are similar to a selected image.
SUMMARY OF THE INVENTION
The present invention is directed to a system, method, and computer-readable medium for finding images in an image catalog that are similar to a selected image based on color or shape. Shape similarity is determined by comparing a wavelet signature derived from the selected image to a wavelet signature derived from each image in the image catalog. Color similarity is determined by comparing the color signature of the selected image to the color signature of each image in the image catalog. A match value, or metric, is assigned for each of these comparisons. The images with the highest match value, i.e., closest match in either color or wavelet, i.e., shape, are returned to the user.
In accordance with other aspects of this invention, the wavelet signature is determined by: centering the image on a field; converting the image to a standard color space; normalizing the colors; performing a wavelet decomposition; determining coefficients for each color plane; and encoding the coefficients based on row/column and sign.
In accordance a further aspect of this invention, YIQ, a color model used in broadcast television, is the standard color space used for determining the wavelet signature.
In accordance with yet a further aspect of the present invention, the colors are normalized, preferably to vary from 0 to 1.
In accordance with still another aspect of this invention, the standard Haar transform is used for the wavelet decomposition.
In accordance with yet another aspect of this invention, 40 of the largest coefficients are stored for each color plane. Preferably, coefficients below a threshold value are not stored.
In accordance with other aspects of the present invention, the color signature for an image is determined by: blurring the image; converting the image to a standard color space; accumulating color pixels in buckets; computing pixel coverage for each of the color buckets; and encoding and storing coverage values for the colors with most coverage.
In accordance with further aspects of this invention, the color space used in determining the color signature is HSV (based on Hue, Saturation and Value elements).
In accordance with yet further aspects of this invention, coverage values are stored for the five colors with the most coverage. Preferably only colors with coverages above a threshold value are stored.
REFERENCES:
patent: 5579471 (1996-11-01), Barber et al.
patent: 5652881 (1997-07-01), Takahashi et al.
patent: 5802361 (1998-09-01), Wang et al.
patent: 5852823 (1998-12-01), De Bonet
patent: 5915250 (1999-06-01), Jain et al.
patent: 6240423 (2001-05-01), Hirata
Bhatnagar Anand
Christensen O'Connor Johnson & Kindness PLLC
Mancuso Joseph
Microsoft Corporation
LandOfFree
Method and system for searching for images based on color... 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 searching for images based on color..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for searching for images based on color... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2931167