Patch-based texture histogram coding for fast image similarity search

1. A method comprising:

dividing a digital image into a plurality of equal size patches;

dividing each patch of the plurality of equal size patches into eight texture directions;

creating a feature vector for each patch of the plurality of equal size patches;

determining a dominant gradient direction corresponding to each patch of the plurality of equal size patches;

generating a texture histogram for each patch of the plurality of equal size patches based on the feature vector for each patch; and

combining the histograms to create a combined histogram to represent the digital image.

2. The method of claim 1 , wherein the dividing is into 8 by 8 (64) equal size patches.

3. The method of claim 1 , wherein the creating uses one or more of the following as features: mean gray, color histogram, and texture histogram.

4. The method of claim 1 , wherein the generating the texture histogram for each patch is performed without regard to color changes.

5. The method of claim 1 further comprising creating a hash code particular to the image based on Principle Component Analysis.

6. The method of claim 1 further comprising measuring similarity of the image with other images, using distance functions.

7. The method of claim 4 , wherein distance functions include one or more of the following: Minkowski metrics, histogram cosine distance, and fuzzy logic.

8. A computer-readable memory including instructions that are executable by one or more processors to perform acts comprising:

dividing a digital image into a plurality of patches, each patch of the plurality of patches having a same size;

dividing each patch of the plurality of patches into eight texture directions;

identifying a dominant gradient direction for each patch of the plurality of patches;

creating a feature vector for each patch of the plurality of patches;

generating a texture histogram for each patch of the plurality of patches based on each feature vector; and

combining the texture histograms to create a combined histogram to represent the digital image.

9. The computer-readable memory of claim 8 , wherein each feature vector is a 64-dimension feature vector.

10. The computer-readable memory of claim 9 , further comprising reducing each 64-dimension feature vector into a 32-dimension feature vector.

11. The computer-readable memory of claim 8 , wherein creating the feature vectors comprises:

quantizing dimensions of the feature vectors into a binary value; and

determining whether the dimensions are above a mean value.

12. The computer-readable memory of claim 8 , wherein the feature vectors are associated with features that are unaffected by patch rotation, image scale change, and color change.

13. The computer-readable memory of claim 8 , further comprising determining a dominant gradient direction of the image.

14. The computer-readable memory of claim 13 , further comprising calculating a hash code based on the combined histogram.

15. A system comprising:

a memory to store instructions; and

one or more processors to execute the instructions to perform acts comprising:

dividing a digital image into equal size patches;

creating a feature vector for each of the patches;

determining a dominant gradient direction corresponding to each of the patches;

generating texture histograms for each of the patches based on each feature vector;

combining the texture histograms using a coding scheme to create a hash code that represents the digital image; and

comparing the hash code to additional hash codes associated with additional images.

16. The system of claim 15 , further comprising identifying duplicate image and near duplicate images based on the comparing.

17. The method of claim 15 , wherein the coding scheme includes projecting the features to a subspace.

18. The method of claim 17 , wherein the subspace is represented by a compact hash code.

19. The method of claim 15 , wherein the coding scheme includes a vector quantization.