Methods and apparatus for video completion

• US 9,013,634 B2
• Filed: 11/24/2010
• Issued: 04/21/2015
• Est. Priority Date: 09/14/2010
• Status: Active Grant

• Current Assignee: Adobe Inc.
• Original Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Inventors: Agarwala, Aseem O., Goldman, Daniel, Leventhal, Daniel H.
• Primary Examiner(s): TORRENTE, RICHARD T
• Application Number:
  US12/954,445
• Publication Number:
  US 20130128121A1
• Time in Patent Office: 1,609 Days
• Field of Search: 348/701
• US Class Current: 348/701
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20182 Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
• CPC Class Codes: G06T 2207/30241 Trajectory
• CPC Class Codes: G06T 2207/30244 Camera pose
• CPC Class Codes: G06T 3/18 Image warping, e.g. rearranging pixels individually
• CPC Class Codes: G06T 5/77 Retouching; Inpainting; Scratch removal
• CPC Class Codes: G06T 7/246 using feature-based methods, e.g. the tracking of corners or segments
• CPC Class Codes: G11B 27/031 Electronic editing of digitised analogue information signals, e.g. audio or video signals
• CPC Class Codes: H04N 13/189 Recording image signals; Reproducing recorded image signals
• CPC Class Codes: H04N 13/221 using the relative movement between cameras and objects

Abstract

Methods, apparatus, and computer-readable storage media for video completion that may be applied to restore missing content, for example holes or border regions, in video sequences. A video completion technique applies a subspace constraint technique that finds and tracks feature points in the video, which are used to form a model of the camera motion and to predict locations of background scene points in frames where the background is occluded. Another frame where those points were visible is found, and that frame is warped using the predicted points. A content-preserving warp technique may be used. Image consistency constraints may be applied to modify the warp so that it fills the hole seamlessly. A compositing technique is applied to composite the warped image into the hole. This process may be repeated until the missing content is filled on all frames.

Claims

1. A method, comprising:

obtaining an input video sequence of a scene comprising a plurality of frames and an indication of an object to be removed from the frames of the input video sequence;

tracking feature points in the input video sequence to generate feature tracks;

factoring the feature tracks into a low-dimensional subspace;

generating a prediction of background scene motion according to the low-dimensional subspace; and

processing a frame of the input video sequence designated as a target frame to fill a region in the target frame occluded by the indicated object with background content from unoccluded regions of one or more other frames of the input video sequence designated as source frames according to the prediction of background scene motion, the occluded region in the target frame is filled using one said unoccluded region from one said source frame at a time such that;

if the one said unoccluded region from the one said source frame is determined to cause an entirety of the occluded region to be filled then compositing the one said source frame to fill the occluded region;

or

if the one said unoccluded region from the one said source frame is determined not to cause the entirety of the occluded region to be filled, then compositing the one said source frame to fill any portion of the occluded region that the background content of the one said source frame is determined to fill, and selecting a next said source frame to fill at least part of a remaining portion of the occluded region.

2. The method as recited in claim 1, further comprising processing each frame in the input video sequence that includes the indicated object.

3. The method as recited in claim 1, wherein processing the target frame according to the prediction of background scene motion comprises:

identifying the unoccluded regions of the one or more source frames according to the prediction of background scene motion; and

warping each source frame to the target frame according to the prediction of background scene motion.

4. The method as recited in claim 3, wherein the warping is performed according to a content-preserving warp technique.

5. The method as recited in claim 3, further comprising applying one or more image consistency energy terms to improve alignment between the target frame and each warped source frame prior to compositing the warped source frame.

6. A system, comprising:

at least one processor; and

a memory comprising program instructions, that are executable by the at least one processor to perform operations comprising;

obtaining an input video sequence of a scene comprising a plurality of frames and an indication of an object to be removed from the frames of the input video sequence;

generating a prediction of background scene motion in the input video sequence; and

processing a frame of the input video sequence designated as a target frame to fill a region in the target frame occluded by the indicated object with background content from unoccluded regions of one or more other frames of the input video sequence designated as source frames, the occluded region in the target frame is filled using one said unoccluded region from one said source frame at a time such that;

if the one said unoccluded region from the one said source frame is determined to cause an entirety of the occluded region to be filled then compositing the one said source frame to fill the occluded region;

or

if the one said unoccluded region from the one said source frame is determined not to cause the entirety of the occluded region to be filled, then compositing the one said source frame to fill any portion of the occluded region that the background content of the one said source frame is determined to fill, and selecting a next said source frame to fill at least part of a remaining portion of the occluded region.

7. The system as recited in claim 6, wherein the processing is performed on each frame in the input video sequence that includes the indicated object.

8. The system as recited in claim 6, wherein each of the one or more source frames is warped to the target frame according to the prediction background scene motion, including applying a content-preserving warp technique.

9. The system as recited in claim 6, wherein one or more image consistency terms are applied to each of the one or more source frames to improve alignment with the target frame, the image consistency energy terms acting to minimize differences in pixel values between a warped one said source frame and the target frame in a band around the region in the target frame occluded by the indicated object.

10. The system as recited in claim 6, wherein generating the prediction of background scene motion in the input video sequence comprises applying a subspace constraint technique that:

finds and tracks feature points in the input video sequence to generate feature tracks;

factors the feature tracks into a low-dimensional subspace; and

generates the prediction of background scene motion according to the low-dimensional subspace.

11. A computer-readable memory storing program instructions that are executable on a computer to perform operations comprising:

obtaining an input video sequence of a scene comprising a plurality of frames and an indication of an object to be removed from the frames of the input video sequence;

generating a prediction of background scene motion in the input video sequence; and

processing a frame of the input video sequence designated as a target frame to fill a region in the target frame occluded by the indicated object with background content from unoccluded regions of one or more other frames designated as source frames according to the prediction of background scene motion, the occluded region in the target frame is filled using one said unoccluded region from one said source frame at a time such that;

if the one said unoccluded region from the one said source frame is determined to cause an entirety of the occluded region to be filled then compositing the one said source frame to fill the occluded region;

or

if the one said unoccluded region from the one said source frame is determined not to cause the entirety of the occluded region to be filled, then compositing the one said source frame to fill any portion of the occluded region that the background content of the one said source frame is determined to fill, and selecting a next said source frame to fill at least part of a remaining portion of the occluded region.

12. The computer-readable memory as recited in claim 11, wherein the processing is performed on each frame in the input video sequence that includes the indicated object.

13. The computer-readable memory as recited in claim 11, wherein each of the one or more source frames is warped to the target frame according to the prediction background scene motion and according to a content-preserving warp technique.

14. The computer-readable memory as recited in claim 11, wherein one or more image consistency terms are applied to each of the one or more source frames to improve alignment with the target frame, the image consistency energy terms acting to minimize differences in pixel values between the source frames and the target frame in a band around the region in the target frame occluded by the indicated object.

15. The computer-readable memory as recited in claim 11, wherein generating the prediction of background scene motion in the input video sequence comprises applying a subspace constraint technique that:

finds and tracks feature points in the input video sequence to generate feature tracks;

factors the feature tracks into a low-dimensional subspace; and

generates the prediction of background scene motion according to the low-dimensional subspace.

16. A computer-readable memory storing program instructions that are executable on a computer to perform operations comprising:

obtaining an input video sequence of a scene comprising a plurality of frames that have been cropped;

tracking feature points in the input video sequence to generate feature tracks;

factoring the feature tracks into a low-dimensional subspace; and

generating the prediction of background scene motion according to the low-dimensional subspace; and

processing each frame of the input video sequence designated as a target frame to fill a cropped region in the target frame with content from background regions of one or more other frames of the input video sequence designated as source frames according to the prediction of background scene motion, the cropped region in the target frame is filled using one said background region from one said source frame at a time such that;

if the one said background region from the one said source frame is determined to cause an entirety of the cropped region to be filled then compositing the one said source frame to fill the cropped region;

or

if the one said background region from the one said source frame is determined not to cause the entirety of the cropped region to be filled, then compositing the one said source frame to fill any portion of the cropped region that the content from the background regions of the one said source frame is determined to fill, and selecting a next said source frame to fill at least part of a remaining portion of the cropped region.

17. The computer-readable memory as recited in claim 16, wherein processing the target frame, comprises:

identifying the background regions of the one or more source frames according to the prediction of background scene motion; and

warping each source frame to the target frame according to the prediction of background scene motion.

18. The computer-readable memory as recited in claim 17, wherein the warping is performed according to a content-preserving warp technique.

19. The computer-readable memory as recited in claim 17, wherein the operations further comprise applying one or more image consistency energy terms to improve alignment between the target frame and each warped source frame prior to said compositing.

20. The computer-readable memory as recited in claim 16, wherein the input video sequence is a stabilized output video sequence generated according to a video stabilization technique that generates the stabilized output video sequence from an original video sequence.