Locality-constrained linear coding systems and methods for image classification

1. A method for classifying an input image, comprising:

detecting one or more feature points on the input image;

extracting one or more descriptors from each feature point;

applying a codebook to quantize each descriptor and generate code from each descriptor;

applying spatial pyramid matching to generate histograms; and

concatenating histograms from all sub-regions to generate a final representation of the image for classification.

2. The method of claim 1 , wherein the final representation of the image for classification comprises a feature vector.

3. The method of claim 1 , wherein the descriptor comprises a SIFT descriptor or a color moment descriptor.

4. The method of claim 1 , wherein each code has only one non-zero element if hard vector quantization (VQ) is used.

5. The method of claim 1 , wherein a small group of elements can be non-zero for soft VQ.

6. The method of claim 1 , wherein multiple codes from inside each sub-region are pooled together by averaging and normalizing into a histogram.

7. The method of claim 1 , comprising performing a fast approximated LLC by first performing a K-nearest-neighbor search and then solving a constrained least square fitting problem.

8. The method of claim 1 , wherein the LLC utilizes locality constraints to project each descriptor into a local-coordinate system, and the projected coordinates are integrated by max pooling to generate a final representation.

9. The method of claim 1 , comprising reconstructing input x i with codebook B i where

c

*

=

arg

⁢

⁢

min

c

⁢



x

i

-

c

i

T

⁢

B



2

+



d

i

·

c

i



2

st

.

⁢

∑

j

M

⁢

⁢

c

j

=

1.

10. The method of claim 1 , comprising:

finding K-Nearest Neighbors of data x i , denoted as B i

reconstructing x i using B i

determining

c

*

=

argmin

c

⁢



x

i

-

c

i

T

⁢

B

i



2

generating c i as an M×1 vector with K non-zero elements whose values are the corresponding c*.

11. An image classifier, comprising:

means for detecting one or more feature points on the input image;

means for extracting one or more descriptors from each feature point

means for applying a codebook to quantize each descriptor and generate code from each descriptor;

means for applying spatial pyramid matching to generate histograms; and

means for concatenating histograms from all sub-regions to generate a final representation of the image for classification.

12. The image classifier of claim 11 , wherein the final representation of the image for classification comprises a feature vector.

13. The image classifier of claim 11 , wherein the descriptor comprises a SIFT descriptor or a color moment descriptor.

14. The image classifier of claim 11 , wherein each code has only one non-zero element if hard vector quantization (VQ) is used.

15. The image classifier of claim 11 , wherein a small group of elements can be non-zero for soft VQ.

16. The image classifier of claim 11 , wherein multiple codes from inside each sub-region are pooled together by averaging and normalizing into a histogram.

17. The image classifier of claim 11 , comprising means for performing a fast approximated LLC by first performing a K-nearest-neighbor search and then solving a constrained least square fitting problem.

18. The image classifier of claim 11 , wherein the LLC utilizes locality constraints to project each descriptor into a local-coordinate system, and the projected coordinates are integrated by max pooling to generate a final representation.

19. The image classifier of claim 11 , comprising means for reconstructing input x i with codebook B i where

c

*

=

arg

⁢

⁢

min

c

⁢



x

i

-

c

i

T

⁢

B



2

+



d

i

·

c

i



2

st

.

⁢

∑

j

M

⁢

⁢

c

j

=

1.

20. The image classifier of claim 11 , comprising:

means for finding K-Nearest Neighbors of data x i , denoted as B i

means for reconstructing x i using B i

where

c

*

=

argmin

c

⁢



x

i

-

c

i

T

⁢

B

i



2

means for generating c i as an M×1 vector with K non-zero elements whose values are the corresponding c*.