System and method for saliency map generation

• US 8,675,966 B2
• Filed: 09/29/2011
• Issued: 03/18/2014
• Est. Priority Date: 09/29/2011
• Status: Expired due to Fees

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Tang, Hao
• Primary Examiner(s): Bali, Vikkram
• Application Number:
  US13/248,947
• Publication Number:
  US 20130084013A1
• Time in Patent Office: 901 Days
• Field of Search: None
• US Class Current: 382/195
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

A system and a method are disclosed for generating a saliency map of an image. The method includes receiving image data representative of image forming elements of an image and determining saliency values for image forming elements by an iterative method. The iterative method includes computing a norm of the image data, computing values of deviation from the norm of the image data of the image forming elements, identifying the image forming elements corresponding to the image data having magnitudes of deviation that meet a pre-determined condition, assigning saliency values to the identified image forming elements based on the values of deviation, and repeating the computing the norm and deviation, identifying image forming elements and assigning saliency values using the image data of image forming elements that have no assigned saliency value. A saliency map of the image based on the assigned saliency values.

Claims

1. A system, comprising:

a memory storing computer-executable instructions; and

a processor coupled to the memory, to execute the instructions, and based at least in part on the execution of the instructions, to;

receive image data representative of image forming elements of an image;

determine saliency values for image forming elements by an iterative method comprising;

i) computing a norm of the image data;

ii) computing values of deviation from the norm of the image data of the image forming elements;

iii) identifying the image forming elements corresponding to the image data having magnitudes of deviation that meet a pre-determined condition;

iv) assigning saliency values to the identified image forming elements based on the values of deviation; and

v) repeating (i)-(iv) using the image data of image forming elements that have no assigned saliency value; and

generate a saliency map of the image based on the assigned saliency values.

2. The system of claim 1, wherein the image forming elements correspond to pixels, superpixels, or image segments;

and wherein the image data are pixel values associated with the pixels, superpixels or image segments, respectively.

3. The system of claim 1, further comprising instructions to compute the norm as a mean, a mode or a median of the values of the image data.

4. The system of claim 1, further comprising instructions to compute the deviation as a Euclidean distance between the values of the image data and the norm.

5. The system of claim 1, wherein, in each iteration, the pre-determined condition is magnitudes of deviation above a threshold value such that a fixed number of the image forming elements are identified in (iii).

6. The system of claim 1, wherein, in each iteration, the pre-determined condition is magnitudes of deviation above a threshold value such that a fixed percentage of images forming elements are identified in (iii).

7. The system of claim 1, wherein, in each iteration, the pre-determined condition is magnitudes of deviation above a threshold value that is defined as a fixed percentage of the maximum value of deviation computed in (ii).

8. The system of claim 1, wherein the saliency value assigned to each image forming element identified in (iii) in each iteration is the respective value of deviation computed for the image forming element.

9. The system of claim 1, wherein the saliency value assigned to the image forming elements identified in (iii) in each iteration is a mean, a mode or a median of the values of deviation computed for the identified image forming elements.

10. The system of claim 1, wherein (i)-(iv) are repeated until about 75%, about 80%, about 90%, or about 95% of the image forming elements have been assigned saliency values.

11. The system of claim 1, wherein the instructions to generate the saliency map of the image further comprises instructions to apply a Gaussian kernel multiplier to the assigned saliency values, and to generate the saliency map of the image based on the weighted saliency values.

12. A system, comprising:

a memory storing computer-executable instructions; and

a processor coupled to the memory, to execute the instructions, and based at least in part on the execution of the instructions, to;

receive image data representative of image forming elements of an image;

determine saliency values for image forming elements by an iterative method comprising;

i) computing a norm of the image data;

ii) computing values of deviation from the norm of the image data of the image forming elements;

iii) identifying the image forming element corresponding to the image data having a highest magnitude of deviation from the norm;

iv) assigning a saliency value to the identified image forming element based on the respective value of deviation; and

v) repeating (i)-(iv) using the image data of image forming elements that have no assigned saliency value; and

generate a saliency map of the image based on the assigned saliency values.

13. The system of claim 12, wherein the image forming elements correspond to pixels, superpixels, or image segments;

and wherein the image data are pixel values associated with the pixels, superpixels or image segments, respectively.

14. The system of claim 12, further comprising instructions to compute the norm as a mean, a mode or a median of the values of the image data.

15. The system of claim 12, further comprising instructions to compute the deviation as a Euclidean distance between the values of the image data and the norm.

16. The system of claim 12, wherein the saliency value assigned to each image forming element identified in (iii) in each iteration is the respective value of deviation computed for the image forming element.

17. The system of claim 12, wherein the instructions to generate the saliency map of the image further comprises instructions to apply a Gaussian kernel multiplier to the assigned saliency values, and to generate the saliency map of the image based on the weighted saliency values.

19. The method of claim 17, wherein the image forming elements correspond to pixels, superpixels, or image segments;

and wherein the image data are pixel values associated with the pixels, superpixels or image segments, respectively.

20. The method of claim 17, further comprising computing the norm as a mean, a mode or a median of the values of the image data.

21. The method of claim 17, wherein the saliency value assigned to each image forming element identified in (iii) in each iteration is the respective value of deviation computed for the image forming element.

22. The method of claim 17, wherein, in each iteration, the pre-determined condition is values of deviation above a threshold value such that a fixed number of the image forming elements are identified in (iii).

23. The method of claim 17, wherein, in each iteration, the pre-determined condition is values of deviation above a threshold value such that a fixed percentage of images forming elements are identified in (iii).

18. A method performed by a computing system comprising at least one processor, the method comprising:

receiving, using the at least one processor, image data representative of image forming elements of an image;

determining, using the at least one processor, a saliency value for image forming elements by an iterative method comprising;

i) computing a norm of the image data;

ii) computing values of deviation from the norm of the image data of the image forming elements;

iii) identifying the image forming elements corresponding to the image data having values of deviation that meet a pre-determined condition;

iv) assigning saliency values to the identified image forming elements based on the values of deviation; and

v) repeating (i)-(iv) using the image data of image forming elements that have no assigned saliency value; and

generating, using the at least one processor, a saliency map of the image based on the assigned saliency values.