Robust and efficient image identification

1. Apparatus for matching a query image against a catalog of images, comprising:

a feature extraction unit configured to:

create a plurality of scaled octave versions of the query image; and,

select a predetermined number of principle features from local intensity extrema in the plurality of scaled octave versions of the query image;

a relationship unit configured to:

establish relationships between a given principle feature and other features in the query image; and

add the relationships as relationship information alongside said principle features; and

a first comparison unit configured to compare principle features and associated relationship information of the query image with principle features and associated relationship information of images of the catalog to find candidate matches.

2. Apparatus according to claim 1 , wherein said feature extraction unit is operative to select said principle features based on points of said image normalized using eigen values.

3. Apparatus according to claim 1 , wherein the feature extraction unit is further configured to find the principle features in blurred level versions of the query image.

4. Apparatus according to claim 1 wherein said relationships are intensities in parts of the image surrounding the principle feature.

5. Apparatus according to claim 4 , wherein said relationships comprise relative sizes.

6. Apparatus according to claim 5 , wherein said relationships are stored in a multi-dimensional vector.

7. Apparatus according to claim 1 , further comprising a second comparison unit operative to compare said query image with said candidate matches to find a nearest neighbour.

8. Apparatus according to claim 7 wherein said second comparison unit is operative to find said nearest neighbour by calculating a transform between said query image and a candidate match.

9. Apparatus according to claim 7 , wherein said second comparison unit is operative to find said nearest neighbour using a Hamming distance.

10. Apparatus according to claim 1 , wherein said feature extractor is operative to obtain said principle features by extracting features from an image and normalizing said extracted features to a set of a predetermined size based on a size of an associated eigen value.

11. Apparatus according to claim 1 , wherein said relationship unit is operative to store said relationship information as a multi-dimensional vector associated with said given principle feature.

12. Apparatus according to claim 11 , wherein said first comparison unit is operative to compare all relationship information of a given principle feature of a query image with a subset of relationship information of a principle feature of a catalogue image, thereby to allow recognition of said query image even when said query image is a cropped version of said catalogue image.

13. Apparatus for matching a query image against a catalog of images, comprising:

a masking unit configured to distinguish between a Laplacian distribution as characterizing image areas and a sparse distribution characterizing text areas, and configured to mask out the text areas;

a feature extraction unit operative for extracting principle features from the query image;

a relationship unit operative for establishing relationships between a given principle feature and other features in the image, and adding the relationships as relationship information alongside the principle features; and,

a first comparison unit operative for comparing principle features and associated relationship information of the query image with principle features and associated relationship information of images of the catalog to find candidate matches.

14. Apparatus according to claim 13 , wherein said relationship unit is configured to divide an image part around a principle feature into overlapping patches, each patch being associated with separate relationship information.

15. A method, implemented by a computing system including a processor and a memory, for matching a query image against a catalog of images, the method comprising:

extracting principle features from the query image, wherein the extracting includes:

creating a plurality of scaled octave versions of the query image; and

selecting a predetermined number of principle features from local intensity extrema in the plurality of scaled octave versions of the query image;

establishing relationships between a given principle feature and other features in the image, and adding the relationships as relationship information alongside the principle features; and

comparing principle features and associated relationship information of the query image with principle features and associated relationship information of images of the catalog to find candidate matches.

16. The method of claim 15 , wherein said relationships are intensities in parts of the image surrounding the principle feature.

17. The method of claim 15 , wherein said relationships comprise relative distances.

18. Apparatus for matching a query image against a catalog of images, comprising:

a feature extraction unit configured to extract principle features from the query image;

a relationship unit configured to divide an image part around a given principle feature into overlapping patches, each patch being associated with separate relationship information for the principle feature; and

a first comparison unit configured to compare principle features and associated relationship information of the query image with principle features and associated relationship information of images of the catalog to find candidate matches.