Detection of demarcating segments in video
A method of detecting frames in a video that demarcate a pre-determined type of video segment within the video is provided. The method includes identifying visually distinctive candidate marker frames within the video, grouping the candidate marker frames into a plurality of groups based on visual similarity, computing a collective score for each of the groups based on temporal proximity of each of the candidate marker frames within the group to related events occurring within the video, and selecting at least one of the groups based on the collective proximity scores as marker frames that demarcate the pre-determined type of video segment. A video processing electronic device and at least one non-transitory computer readable storage medium having computer program instructions stored thereon for performing the method are also provided.
1. A method of detecting frames in a video that demarcate a pre-determined type of video segment within the video, comprising the steps of:
identifying, based at least on a probability distribution for a plurality of visual features of the frames in the video, candidate marker frames within the video, wherein each of the candidate marker frames has a visual feature set, and wherein each of the candidate marker frames is visually distinctive based on dissimilarity of the visual feature set to a statistical model of the video;
identifying events within the video as input video events;
grouping the candidate marker frames into a plurality of groups based on visual similarity to one another;
computing a collective score for each of the groups based on temporal proximity of each of the candidate marker frames in each of the groups to input video events; and
selecting at least one of the groups based on the collective scores of marker frames that demarcate the pre-determined type of video segment.
2. A method according to claim 1 , further comprising the step of locating the pre-determined type of video segments within the video by detecting location of marker frames in the video.
3. A method according to claim 2 , further comprising the step of creating non-linear video content including the pre-determined type of video segments.
4. A method according to claim 1 , wherein said input video events within the video are detected events within the video.
5. A method according to claim 4 , wherein said detected events are selected from a group consisting of slow motion, telestration, and absence of score panel, and wherein said selecting step includes determining the group of candidate markers having a collective score indicating the closest temporal proximity to the detected events.
6. A method according to claim 1 , wherein said input video events within the video are candidate marker frames from a second group of candidate marker frames, and wherein each of the candidate marker frames from the second group of candidate marker frames demarcates the end of a pre-determined type of video segment within the video.
7. A method according to claim 1 , wherein said identifying step includes a step of separately evaluating each frame of the video.
8. A method according to claim 7 , wherein said evaluating step includes transforming each frame of the video to obtain for each frame a set of visual features.
9. A method according to claim 8 , wherein the set of visual features includes at least one of a Color Layout Descriptor (CLD), a color histogram, and an Edge Histogram Descriptor (EHD).
10. A method according to claim 8 , wherein said identifying step includes a step of generating the probability distribution for the set of visual features.
11. A method according to claim 10 , wherein said identifying step includes classifying frames of the video as being a visually distinctive candidate marker frame based on crossing a threshold defined within the probability distribution.
12. A method according to claim 1 , wherein said grouping step includes a step of determining visual similarity between candidate marker frames so that visually similar candidate marker frames are placed in the same group.
13. A method according to claim 12 , wherein said grouping step includes hashing.
14. A method according to claim 1 , wherein the marker frames demarcate at least one of the beginning and end of the pre-determined type of video segment.
15. A method according to claim 1 , wherein the pre-determined type of video segment includes replay video segments.
16. A method according to claim 1 , further comprising the step of searching and matching for the marker frames and neighbor frames surrounding the marker frames at different locations within the video.
17. A video processing electronic device for detecting frames in a video that demarcate a pre-determined type of video segment within the video, comprising at least one processing unit configured to:
identify, based at least on a probability distribution for a plurality of visual features of the frames in the video, candidate marker frames within the video, wherein each of the candidate marker frames has a visual feature set, and wherein each of the candidate marker frames is visually distinctive based on dissimilarity of the visual feature set to a statistical model of the video;
identify events within the video as input video events;
group the candidate marker frames into a plurality of groups based on visual similarity;
compute a score for each of the groups based on temporal proximity of each of the candidate marker frames within the group to input video events in the pre-determined type of video segment; and
select at least one of the groups based on the score of marker frames that demarcate the pre-determined type of video segment.
18. A video processing electronic device according to claim 17 , wherein said at least one processing unit is configured to automatically locate the pre-determined type of video segments within the video by detecting the marker frames in the video and to create video content including the pre-determined video segments.
19. At least one non-transitory computer readable storage medium having computer program instructions stored thereon that, when executed by at least one processor, cause the at least one processor to automatically detect frames in a video that demarcate a pre-determined type of video segment within the video by performing the following operations:
identify, based at least on a probability distribution for a plurality of visual features of the frames in the video, candidate marker frames within the video, wherein each of the candidate marker frames has a visual feature set, and wherein each of the candidate marker frames is visually distinctive based on dissimilarity of the visual feature set to a statistical model of the video;
identify events within the video as input video events;
group the candidate marker frames into a plurality of groups based on visual similarity;
compute a score for each of the groups based on temporal proximity of each of the candidate marker frames within the group to input video events in the pre-determined type of video segment; and
select at least one of the groups based on the score of marker frames that demarcate the pre-determined type of video segment.
20. At least one non-transitory computer readable storage medium having computer program instructions stored thereon according to claim 19 , wherein when the computer program instructions are executed by at least one processor, the pre-determined type of video segments are automatically detected within the video by detecting the marker frames in the video.