Image fusion using sparse overcomplete feature dictionaries

• US 9,152,881 B2
• Filed: 09/13/2013
• Issued: 10/06/2015
• Est. Priority Date: 09/13/2012
• Status: Active Grant

• Current Assignee: TRIAD National Security, LLC
• Original Assignee: Los Alamos National Security LLC (Government of the United States of America)
• Inventors: Brumby, Steven P., Bettencourt, Luis, Kenyon, Garrett T., Chartrand, Rick, Wohlberg, Brendt
• Primary Examiner(s): Patel, Nirav G
• Assistant Examiner(s): LEMIEUX, IAN L
• Application Number:
  US14/026,295
• Publication Number:
  US 20140072209A1
• Time in Patent Office: 753 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 18/21345 enforcing sparsity or involving a domain transformation
• CPC Class Codes: G06F 18/2321 using statistics or function optimisation, e.g. modelling of probability density functions
• CPC Class Codes: G06V 10/513 Sparse representations

Abstract

Approaches for deciding what individuals in a population of visual system “neurons” are looking for using sparse overcomplete feature dictionaries are provided. A sparse overcomplete feature dictionary may be learned for an image dataset and a local sparse representation of the image dataset may be built using the learned feature dictionary. A local maximum pooling operation may be applied on the local sparse representation to produce a translation-tolerant representation of the image dataset. An object may then be classified and/or clustered within the translation-tolerant representation of the image dataset using a supervised classification algorithm and/or an unsupervised clustering algorithm.

Claims

1. A computer-implemented method, comprising:

learning, by a computing system, a sparse overcomplete feature dictionary for classifying and/or clustering a remote sensing image dataset;

building, by the computing system, a local sparse representation of the image dataset using the learned sparse overcomplete feature dictionary; and

applying, by the computing, system, a local maximum pooling operation on the local sparse representation to produce a translation-tolerant representation of the image dataset.

2. The computer-implemented method of claim 1, wherein the learning of the sparse overcomplete feature dictionary comprises:

initializing, by the computing system, atoms φ

_k of a feature dictionary Φ

either by imprinting a set of unlabeled patches x, or by initializing φ

_k using random vectors;

for each unlabeled patch in x, seeking, by the computing system, a coefficient vector y such that y is sparse and Φ

y approximates x;

finding, by the computing system, an approximate solution for y; and

updating Φ

, by the computing system, using a learning rule.

3. The computer-implemented method of claim 1, wherein the building of the local sparse representation of the image dataset comprises applying a greedy matching pursuit algorithm or a non-convex basis pursuit algorithm.

4. The computer-implemented method of claim 1, further comprising:

down-sampling the image dataset, by the computing system, prior to building the local sparse representation of the image dataset.

5. The computer-implemented method of claim 1, further comprising:

classifying and/or clustering, by the computing system, an object within the translation-tolerant representation of the image dataset using a supervised classification algorithm and/or an unsupervised clustering algorithm.

6. The computer-implemented method of claim 5, wherein the classifying and/or clustering of the object comprises categorizing the translation-tolerant representation of the image dataset in a predetermined number of categories.

7. The computer-implemented method of claim 6, wherein the categorizing of the translation-tolerant representation of the image dataset comprises applying a cluster model to the translation-tolerant representation of the image dataset.

8. The computer-implemented method of claim 6, further comprising:

augmenting, by the computing system, a predetermined number of background categories with a number of target categories using the sparse overcomplete feature dictionary and supervised clustering.

9. The computer-implemented method of claim 8, further comprising:

classifying and/or clustering the image dataset, by the computing system, in response to a minimum distance algorithm applied across the augmented categories.

10. A computer program embodied on a non-transitory computer-readable medium, the program configured to cause at least one processor to:

initialize atoms φ

_k of a feature dictionary Φ

by imprinting a set of unlabeled patches x;

for each unlabeled patch in x, seek a coefficient vector y such that y is sparse and Φ

y approximates x;

find an approximate solution for y; and

update Φ

using a learning rule.

11. The computer program of claim 10, wherein the program is further configured to cause the at least one processor to find the approximate solution of y using a simple matching pursuit algorithm, orthogonal matching pursuits, an l−

1 convex basis pursuit, or a non-convex basis pursuit.

12. The computer program of claim 10, wherein the program is further configured to cause the at least one processor to draw the set of unlabeled patches x for an image dataset.

13. The computer program of claim 12, wherein the learning rule is defined by:

∀

φ

_k∈

Φ

,Δ

φ

_k=η

y_k(x−

Φ

y)

where η

is a learning rate and k is a number of indexes over φ

_k in Φ

.

14. An apparatus, comprising:

memory storing computer program instructions; and

at least one processor configured to execute the computer program instructions, the at least one processor configured to;

learn a sparse overcomplete feature dictionary for an image dataset,

build a local sparse representation of the image dataset using the learned feature dictionary,

apply a local maximum pooling operation on the local sparse representation to produce a translation-tolerant representation of the image dataset, and

classify and/or cluster an object within the translation-tolerant representation of the image dataset using a supervised classification algorithm and/or an unsupervised clustering algorithm.

15. The apparatus of claim 14, wherein the building of the local sparse representation of the image dataset comprises applying a greedy matching pursuit algorithm or a non-convex basis pursuit noisy algorithm.

16. The apparatus of claim 14, wherein the at least one processor is further configured to down-sample the image dataset prior to building the local sparse representation of the image dataset.

17. The apparatus of claim 14, wherein the detecting of the object comprises categorizing the translation-tolerant representation of the image dataset in a predetermined number of categories.

18. The apparatus of claim 17, wherein the at least one processor is further configured to:

learn a number of categories of objects in the image dataset by using supervised and unsupervised clustering of sparse image region representations in the sparse overcomplete feature dictionary; and

classify the image dataset in response to a minimum distance algorithm applied across the learned object categories.