APPARATUS, METHOD, AND MEDIUM OF GENERATING VISUAL ATTENTION MAP

• US 20100246954A1
• Filed: 03/18/2010
• Published: 09/30/2010
• Est. Priority Date: 03/31/2009
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: KIM, Ji Won, Jung, Yong Ju
• Granted Patent: US 8,406,524 B2
• Time in Patent Office: Days
• US Class Current: 382/170
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 10/56 relating to colour

Abstract

A method, apparatus, and medium of generating a visual attention map. A visual attention map to extract visual attention may be generated to convert a two-dimensional (2D) image into a three-dimensional (3D) image based on visual attention. The 2D image may be downscaled and at least one downscaled image may be generated. A feature map may be extracted from the 2D image and the at least one downscaled image, and the visual attention map may be generated.

Claims

1. An apparatus of generating a visual attention map, the apparatus comprising:

a downscaling unit to downscale a two-dimensional (2D) image and to generate at least one downscaled image;

a feature map generation unit to extract feature information from the 2D image and the at least one downscaled image, and to generate at least one feature map; and

an attention computation unit to perform low-level attention computation using the at least one feature map, and to generate the visual attention map based on a result of the low-level attention computation.

2. The apparatus of claim 1, wherein the downscaling unit downscales the 2D image into a plurality of rectangular blocks.

3. The apparatus of claim 1, wherein the downscaling unit comprises:

a horizontal downscaling unit to downscale the 2D image in a horizontal direction, and to generate the at least one downscaled image; and

a vertical downscaling unit to downscale the 2D image in a vertical direction, and to generate the at least one downscaled image.

4. The apparatus of claim 1, wherein the attention computation unit comprises:

an area setting unit to set a center area and a surround area with respect to the at least one feature map, the surround area including the center area;

a histogram computation unit to compute a feature information histogram distance between the center area and the surround area; and

an attention map generation unit to generate the visual attention map using the feature information histogram distance

5. The apparatus of claim 4, wherein feature information histograms of the feature information histogram distance are at least one of an intensity histogram and a color histogram.

6. The apparatus of claim 4, wherein the histogram computation unit computes the feature information histogram distance using a Chi-square of a histogram of the center area and a histogram of the surround area.

7. The apparatus of claim 4, wherein the area setting unit comprises:

a unit block setting unit to set a unit block of a square or a rectangle;

a center area setting unit to set the center area as a size of the unit block; and

a surround area setting unit to set the surround area as a total size of a plurality of unit blocks.

8. The apparatus of claim 1, wherein the attention computation unit comprises:

an area setting unit to set a center area and a surround area with respect to the at least one feature map, the surround area including the center area;

a histogram generation unit to generate respective feature information histograms of the center area and the surround area;

a histogram computation unit to compute a feature information histogram distance using a moment of the feature information histograms; and

an attention map generation unit to generate the visual attention map based on the feature information histogram distance,

wherein the moment includes at least one of a mean, a variance, a standard deviation, and a skewness.

9. An apparatus of generating a visual attention map, the apparatus comprising:

a block division unit to divide a 2D image into a plurality of blocks;

an area setting unit to select at least one block of the plurality of blocks as a center area, and to select at least one block from a surround area adjacent to the center area as a surround block;

a feature information extraction unit to extract first feature information of the center area and second feature information of the surround area; and

a histogram computation unit to compute a histogram distance between the first feature information and the second feature information, and to generate the visual attention map.

10. The apparatus of claim 9, wherein the histogram computation unit computes the histogram distance using a Chi-square of a histogram of the first feature information and a histogram of the second feature information.

11. The apparatus of claim 9, wherein the histogram computation unit generates the visual attention map using a moment of respective histograms of the first feature information and the second feature information, the moment including at least one of a mean, a variance, a standard deviation, and a skewness.

12. A method of generating a visual attention map, the method comprising:

downscaling a 2D image and generating at least one downscaled image;

extracting feature information from the 2D image and the at least one downscaled image, and generating at least one feature map; and

performing low-level attention computation using the at least one feature map, and generating the visual attention map based on a result of the low-level attention computation.

13. The method of claim 12, wherein the generating of the at least one downscaled image comprises:

downscaling the 2D image in a horizontal direction, and generating the at least one downscaled image; and

downscaling the 2D image in a vertical direction, and generating the at least one downscaled image.

14. The method of claim 12, wherein the generating of the visual attention map comprises:

setting a center area and a surround area with respect to the at least one feature map, the surround area including the center area;

computing a feature information histogram distance between the center area and the surround area; and

generating the visual attention map using the feature information histogram distance.

15. The method of claim 14, wherein feature information histograms of the feature information histogram distance are at least one of an intensity histogram and a color histogram.

16. The method of claim 14, wherein the computing of the feature information histogram distance uses a Chi-square of a histogram of the center area and a histogram of the surround area.

17. The method of claim 14, wherein the setting of the center area and the surround area comprises:

setting a unit block of a square or a rectangle;

setting the center area as a size of the unit block; and

setting the surround area as a total size of a plurality of unit blocks.

18. The method of claim 12, wherein the generating of the visual attention map comprises:

setting a center area and a surround area with respect to the at least one feature map, the surround area including the center area;

generating respective feature information histograms of the center area and the surround area;

computing a feature information histogram distance using a moment of the feature information histograms; and

generating the visual attention map based on the feature information histogram distance,

wherein the moment includes at least one of a mean, a variance, a standard deviation, and a skewness.

19. A computer-readable recording medium storing a program causing at least one processing element to implement a method of generating a visual attention map, the method comprising:

downscaling a 2D image and generating at least one downscaled image;

extracting feature information from the 2D image and the at least one downscaled image, and generating at least one feature map; and

performing low-level attention computation using the at least one feature map, and generating the visual attention map based on a result of the low-level attention computation.