Robust and efficient image identification
Apparatus for matching a query image against a catalog of images, comprises: a feature extraction unit operative for extracting principle features from said query image; a relationship unit operative for establishing relationships between a given principle feature and other features in the image, and adding said relationships as relationship information alongside said principle features; and a first comparison unit operative for comparing principle features and associated relationship information of said query image with principle features and associated relationship information of images of said catalog to find candidate matches.
1. A method of analyzing a digital image to determine a descriptor describing the digital image, the method comprising:
creating a plurality of scaled octave versions of the digital image;
selecting a subset of feature points within multiple local extrema, wherein each feature point is selected from the plurality of scaled octave versions of the digital image and has an associated scale octave;
creating a vector for each of the feature points, each feature point vector based on the normalized intensity values of a patch around the feature point,
wherein the patch size is determined by the associated scale octave;
generating, for a particular feature point, a concatenated vector based on a plurality of feature point vectors, wherein the patches associated with each of the plurality of feature point vectors overlap the particular feature point; and
creating a descriptor for the digital image based on the concatenated vector, wherein the descriptor is a compact representation of feature points defining the digital image.
2. The method of claim 1 , further comprising:
generating a scale-space pyramid from a series of scale octaves of the digital image; and
localizing the multiple local extrema in the scale-space pyramid, wherein the multiple local extrema are greater than a predetermined intensity.
3. The method of claim 1 , wherein the concatenated vector is based on five feature point vectors.
4. The method of claim 1 , wherein the descriptor uses a scale space determined by the scale octave associated with the feature point for which the concatenated vector was generated.
5. The method of claim 4 , wherein the descriptor is invariant to scale.
6. The method of claim 1 , further comprising:
constructing a line of demarcation between a first region and a second region,
wherein the demarcation line is determined by edge distributions of the first region and the second region, each of the first and second region differing in intensity; and
defining the digital image to be analyzed based on the demarcation line.
7. The method of claim 6 , wherein the first region is a continuous tone region and the second region is a graphics region.
8. The method of claim 6 , wherein the line of demarcation is a hyperplane or a set of hyperplanes.
9. The method of claim 8 , wherein a distance measured between the hyperplane and a nearest feature point defined in the digital image defines a functional margin.
10. The method of claim 1 , wherein the descriptor is resilient to distortions in a digital image plane.
11. The method of claim 1 , further comprising determining an orthogonal coordinate system based on the largest variance between each feature point vector.
12. The method of claim 1 , further comprising generating a catalog of images, each image having an associated descriptor describing the image.
13. A non-transitory computer-readable medium comprising instructions executable by at least one processor to perform a method of analyzing a digital image to determine a descriptor describing the digital image, the method comprising:
creating a plurality of scaled octave versions of the digital image;
selecting a subset of feature points within multiple local extrema, wherein each feature point is selected from the plurality of scaled octave versions of the digital image and has an associated scale octave;
creating a vector for each of the feature points, each feature point vector based on the normalized intensity values of a patch around the feature point,
wherein the patch size is determined by the associated scale octave;
generating, for a particular feature point, a concatenated vector based on a plurality of feature point vectors, wherein the patches associated with each of the plurality of feature point vectors overlap the particular feature point; and
creating a descriptor for the digital image based on the concatenated vector, wherein the descriptor is a compact representation of feature points defining the digital image.
14. The non-transitory computer-readable medium of claim 13 , the method further comprising:
generating a scale-space pyramid from a series of scale octaves of the digital image; and
localizing the multiple local extrema in the scale-space pyramid, wherein the multiple local extrema are greater than a predetermined intensity.
15. The non-transitory computer-readable medium of claim 13 , wherein the descriptor is invariant to scale.
16. The non-transitory computer-readable medium of claim 13 , wherein the descriptor is resilient to distortions in a digital image plane.
17. The non-transitory computer-readable medium of claim 13 , the method further comprising determining an orthogonal coordinate system based on the largest variance between each feature point vector.
18. The non-transitory computer-readable medium of claim 13 , the method further comprising generating a catalog of images, each image having an associate descriptor describing the image.
19. A system including at least one processor and memory to analyze a digital image to determine a descriptor describing the image, the system comprising:
a feature extraction unit configured to:
create a plurality of scaled octave versions of the digital image;
select a subset of feature points within multiple local extrema, wherein each feature point is selected from the plurality of scaled octave versions of the digital image and has an associated scale octave; and
create a vector for each of the feature points, each feature point vector based on the normalized intensity values of a patch around the feature point,
wherein the patch size is determined by the associated scale octave; and
a relationship unit configured to:
generate, for a particular feature point, a concatenated vector based on a plurality of feature point vectors, wherein the patches associated with each of the plurality of feature point vectors overlap the particular feature point; and
create a descriptor for the digital image based on the concatenated vector, wherein the descriptor is a compact representation of feature points defining the digital image.