Decomposition of a video stream into salient fragments

1. A computer-implemented method comprising:

extracting a plurality of salient fragments of a video, each salient fragment including a subset of frames of the video and a subset of pixels from the subset of frames, wherein a salient fragment is extracted by:

identifying a subset of time-related frames of the video that include salient portions as the subset of frames;

identifying a subset of pixels that are associated with the salient portion in each frame of the subset of time-related frames of the video as the salient fragment; and

storing the salient fragment, including the subset of pixels associated with the salient portion from the subset of time-related frames of the video in a database of the plurality of salient fragments;

extracting non-salient portions of the video that are irrelevant to the plurality of salient fragments;

receiving a query;

retrieving, from the database of the plurality of salient fragments, a first set of salient fragments based on the query by determining a graph connecting the salient fragments based on a spatial-temporal overlap or causality relationship between the salient fragments, identifying a first salient fragment related to the query, and retrieving the salient fragments connected in the graph to the first salient fragment as the first set of salient fragments; and

synthesizing a video composition including the first set of salient fragments and the non-salient portions of the video associated with the first set of salient fragments.

2. The computer-implemented method of claim 1 , further comprising identifying the salient fragment based on at least one of motion detection, contrast, color, and semantic information.

3. The computer-implemented method of claim 1 , wherein each salient fragment is a portion of the video for a single salient activity localized in time and space.

4. The computer-implemented method of claim 1 , wherein the query includes at least one of: the first salient fragment of the plurality of salient fragments, the first salient fragment and a second salient fragment of the plurality of salient fragments, a time interval, or an attribute associated with salient fragments of the plurality of salient fragments.

5. The computer-implemented method of claim 1 , further comprising retrieving a second set of salient fragments from the first set of salient fragments based on a second query.

6. The computer-implemented method of claim 1 , further comprising building an index for the database of the plurality of salient fragments, wherein the index is used for fast retrieval of salient fragments within the database.

7. The computer-implemented method of claim 6 , wherein the index is built from the graph connecting the first salient fragment and a second salient fragment.

8. The computer-implemented method of claim 1 , wherein synthesizing the video composition further comprises determining one or more time shifts that minimizes the spatial-temporal overlap between objects in the salient fragment.

9. The computer-implemented method of claim 1 , further comprising grouping the first set of salient fragments into clusters, wherein the clusters include salient fragments having the spatial-temporal or causal relationship.

10. The computer-implemented method of claim 9 , wherein synthesizing the video composition further comprises generating a synopsis based on time-shifting the clusters.

11. The computer-implemented method of claim 1 , further comprising assigning a weight to the salient fragment in the first set of salient fragments based on an activity level associated with the salient fragment.

12. The computer-implemented method of claim 11 , wherein synthesizing the video composition further comprises assigning the salient fragment to an editing layer based on the weight corresponding to the salient fragment.

13. The computer-implemented method of claim 1 , wherein synthesizing the video composition further comprises performing a spatial-temporal distortion on the first salient fragment of the first set of salient fragments to avoid a collision with a second salient fragment of the first set of salient fragments.

14. The computer-implemented method of claim 13 , further comprising bounding the spatial-temporal distortion by a zone of permissible activity within the non-salient portions of the video.

15. A system comprising:

one or more processors; and

a memory, the memory storing instructions, which when executed by the one or more processors cause the one or more processors to:

extract a plurality of salient fragments of a video, each salient fragment including a subset of frames of the video and a subset of pixels from the subset of frames, wherein a salient fragment is extracted by:

identifying a subset of time-related frames of the video that include salient portions as the subset of frames;

identifying a subset of pixels that are associated with the salient portion in each frame of the subset of time-related frames of the video as the salient fragment; and

storing the salient fragment, including the subset of pixels associated with the salient portion from the subset of time-related frames of the video in a database of the plurality of salient fragments;

extract non-salient portions of the video that are irrelevant to the plurality of salient fragments;

receive a query;

retrieve, from the database of the plurality of salient fragments, a first set of salient fragments based on the query by determining a graph connecting the salient fragments based on a spatial-temporal overlap or causality relationship between the salient fragments, identifying a first salient fragment related to the query, and retrieving the salient fragments connected in the graph to the first salient fragment as the first set of salient fragments; and

synthesize a video composition including the first set of salient fragments and the non-salient portions of the video associated with the first set of salient fragments.

16. The system of claim 15 , wherein the instructions cause the one or more processors to identify the salient fragment based on at least one of motion detection, contrast, color, and semantic information.

17. The system of claim 15 , wherein the query includes at least one of: the first salient fragment of the plurality of salient fragments, the first salient fragment and a second salient fragment of the plurality of salient fragments, a time interval, or an attribute associated with salient fragments of the plurality of salient fragments.

18. The system of claim 15 , wherein the instructions further cause the one or more processors to build an index for the database of the plurality of salient fragments, wherein the index is used for fast retrieval of salient fragments within the database.

19. The system of claim 18 , wherein the index is built from the graph connecting the first salient fragment and a second salient fragment.

20. A computer program product comprising a non-transitory computer readable medium storing a computer readable program, wherein the computer readable program when executed causes a computer to:

extract a plurality of salient fragments of a video, each salient fragment including a subset of frames of the video and a subset of pixels from the subset of frames, wherein a salient fragment is extracted by:

identifying a subset of time-related frames of the video that include salient portions as the subset of frames;

identifying a subset of pixels that are associated with the salient portion in each frame of the subset of time-related frames of the video as the salient fragment; and

storing the salient fragment, including the subset of pixels associated with the salient portion from the subset of time-related frames of the video in a database of the plurality of salient fragments;

extract non-salient portions of the video that are irrelevant to the plurality of salient fragments;

receive a query;

retrieve, from the database of the plurality of salient fragments, a first set of salient fragments based on the query by determining a graph connecting the salient fragments based on a spatial-temporal overlap or causality relationship between the salient fragments, identifying a first salient fragment related to the query, and retrieving the salient fragments connected in the graph to the first salient fragment as the first set of salient fragments; and

synthesize a video composition including the first set of salient fragments and the non-salient portions of the video associated with the first set of salient fragments.