Method and apparatus for generating superpixels

Patent metadata

Field	Value
Publication number	US-9928574-B2
Application number	US-201414542899-A
Country	US
Kind code	B2
Filing date	Nov 17, 2014
Priority date	Nov 19, 2013
Publication date	Mar 27, 2018
Grant date	Mar 27, 2018

Abstract

A method and an apparatus for generating superpixels for a sequence of images are described. A segmenter segments an image of the sequence of images into a set of initial superpixels. A motion analyzer then determines static and non-static areas in the sequence of images. Based on an output of the motion analyzer the segmenter generates updated superpixels for non-static areas of the sequence of images, and maintains those superpixels of the set of initial superpixels belonging to static areas of the sequence of images.

First claim

The invention claimed is:

1. A method for generating superpixels for a sequence of images, the method comprising:
segmenting an image of the sequence of images into a set of initial superpixels;
determining static and non-static areas in the sequence of images by analyzing motion vector means values and their variances within each superpixel;
generating updated superpixels for non-static areas of the sequence of images; and
maintaining those superpixels of the set of initial superpixels belonging to static areas of the sequence of images;
wherein a superpixel is set as immutable if the motion vector mean value is smaller than a first determined value and the variance is smaller than a second determined value, and as mutable else, a superpixel marked as immutable being set to mutable if a directed mean value of a motion vector field divergence value is smaller than a third determined value and a mean motion vector field rotation value within said superpixel is larger than a fourth determined value.

2. The method according to claim 1 , wherein the first threshold is determined from an average size of the superpixels of an image.

3. The method according to claim 1 , wherein the first determined value is determined from an average superpixel size determined by analyzing all superpixels present in an image.

4. The method according to claim 1 , wherein immutable superpixels adjacent to superpixels in static areas are analyzed and the number of mutable superpixels is enlarged in dependence on motion vectors found within the mutable superpixels.

5. The method according to claim 1 , wherein only superpixel seed points belonging to immutable superpixels are frozen, while the segmenting is applied to all superpixels.

6. The method according to claim 1 , wherein superpixel seed points and pixels of immutable superpixels are frozen and the segmenting is applied to pixels of mutable superpixels only.

7. An apparatus configured to generate superpixels for a sequence of images, the apparatus comprising:
a motion analyzer configured to determine static and non-static areas in the sequence of images by analyzing absolute motion vector mean values and their variances within each superpixel; and
a segmenter configured to segment an image of the sequence of images into a set of initial superpixels, generate updated superpixels for non-static areas of the sequence of images, and maintain those superpixels of the set of initial superpixels belonging to static areas of the sequence of images;
wherein a superpixel is set as immutable if the motion vector mean value is smaller than a first determined value and the variance is smaller than a second determined value, and as mutable else, a superpixel marked as immutable being set to mutable if a directed mean value of a motion vector field divergence value is smaller than a third determined value and a mean motion vector field rotation value within said superpixel is larger than a fourth determined value.

8. The apparatus according to claim 7 , wherein the segmenter is configured to determine the first determined value from an average size of the superpixels of an image.

9. The apparatus according to claim 7 , wherein the segmenter is configured to determine the first determined value from an average superpixel size determined by analyzing all superpixels present in an image.

10. The apparatus according to claim 7 , wherein the segmenter is configured to analyze immutable superpixels adjacent to superpixels in static areas and to enlarge the number of mutable superpixels in dependence on motion vectors found within the mutable superpixels.

11. The apparatus according to claim 7 , wherein the segmenter is configured to freeze only superpixel seed points belonging to immutable superpixels, but to apply the segmenting to all superpixels.

12. The apparatus according to claim 7 , wherein the segmenter is configured to freeze superpixel seed points and pixels of immutable superpixels and to apply the segmenting to pixels of mutable superpixels only.

13. A non-transitory computer readable storage medium having stored therein instructions enabling generating superpixels for a sequence of images, which when executed by a computer, cause the computer to:
segment an image of the sequence of images into a set of initial superpixels;
determine static and non-static areas in the sequence of images by analyzing absolute motion vector mean values and their variances within each superpixel;
generate updated superpixels for non-static areas of the sequence of images; and
maintain those superpixels of the set of initial superpixels belonging to static areas of the sequence of images;
wherein a superpixel is set as immutable if the motion vector mean value is smaller than a first determined value and the variance is smaller than a second determined value, and as mutable else, a superpixel marked as immutable being set to mutable if a directed mean value of a motion vector field divergence value is smaller than a third determined value and a mean motion vector field rotation value within said superpixel is larger than a fourth determined value.

14. The non-transitory computer readable storage medium according to claim 13 , wherein the instructions cause the computer to determine the first determined value from an average size of the superpixels of an image.

15. The non-transitory computer readable storage medium according to claim 13 , wherein the instructions cause the computer to determine the first determined value from an average superpixel size determined by analyzing all superpixels present in an image.

16. The non-transitory computer readable storage medium according to claim 13 , wherein the instructions cause the computer to analyze immutable superpixels adjacent to superpixels in static areas and to enlarge the number of mutable superpixels in dependence on motion vectors found within the mutable superpixels.

17. The non-transitory computer readable storage medium according to claim 13 , wherein the instructions cause the computer to freeze only superpixel seed points belonging to immutable superpixels, but to apply the segmenting to all superpixels.

18. The non-transitory computer readable storage medium according to claim 13 , wherein the instructions cause the computer to freeze superpixel seed points and pixels of immutable superpixels and to apply the segmenting to pixels of mutable superpixels only.

Assignees

• Thomson Licensing
• Thompson Licensing Sa

Inventors

• Gandolph Dirk
• Jachalsky Joern
• Putzke-Roeming Wolfram

Classifications

• G06T7/11 Primary

  Region-based segmentation · CPC title

• G06T5/00 Primary

  Image enhancement or restoration · CPC title

• G06T2207/20012

  Locally adaptive · CPC title

• G06T7/194

  involving foreground-background segmentation · CPC title

• G06T2207/20156

  Automatic seed setting · CPC title