A method and system for matching clips with videos via media analysis

A method of identifying portions of a video which comprise overlaid text, comprising:

• in a first phase:
• receiving the video;
• receiving textless clips/video;
• comparing each textless clip to portions of the video;
• identifying those textless clips which match only one portion of the video;
• based on said identification, processing the textless clips and the matched portions of the video by:
• detecting the text images in each, and
• comparing the text images in each matched pair;
• identifying those matched pairs as each having a text box as background text;
• identifying these matched pairs in which only one has a text box as being overlaid text;
• training a model in dependence on said two identifications,
• in a probability step:
• receiving the video;
• detecting portions of the video having text;
• applying the model to said portions to compute a probability that the portion having text has overlaid text;
• thereby providing a probability of overlaid text for each portion having text,
• in a matching step:
• receiving the video;
• receiving the textless clips;
• identifying those textless clips of the video which match more than one portion of the video;
• selecting only one portion for that textless
• clip based on the portion having the highest probability of overlaid text;
• wherein all textless clips for which a match is found are uniquely matched to one portion of the video,
• in a second phase:
• receiving the video;
• receiving textless clips;
• comparing each textless clip to portions of the video.

A method of automatically identifying portions of a video which comprises overlaid text, comprising:

• receiving the video;
• receiving textless clips;
• comparing each textless clip to portions of the video;
• identifying those textless clips which match only one portion of the video;
• matching those textless clips which match only one portion of the video to that portion;
• identifying those textless clips which match more than one portion of the video;
• selecting one portion of the video for that textless clip based on the portion of the video determined to have the highest probability of containing overlaid text;
• match that textless clip to the selected portion of the video;
• wherein all textless clips are matched to one portion of the video.

A method of automatically identifying portions of a full-length video as being associated with overlaid text, comprising:

• receiving the full-length video;
• receiving textless clips of the full-length video;
• receiving a model for identifying portions of the full-length video as comprising overlaid text;
• identifying text in portions of the received full-length video, and in dependence on the model apportioning a probability value to such portion, the probability value indicating the probability of the text being overlaid text;
• identifying the textless clips which match any portion of the full-length video to provide matched pairs;
• if a textless clip matches only one portion of the full-length video, matching that textless clip to that portion of the full-length video;
• if a textless clip matches multiple portions of the full-length video, matching that textless clip to the one of the portions having the highest probability;
• and
• identifying each portion of the full-length video uniquely matched to a textless clip as containing overlaid text.

A method of training a classifier, which classifier is used to process a full-length video to identify portions of the video which contain overlaid text, the method comprising:

• receiving the full-length video;
• receiving textless clips of the full-length video;
• identifying textless clips which match to only one portion of the full-length video to provide matched pairs;
• and
• detecting in each textless clip and portion of full-length video of each matched pair whether text is present;
• wherein if text is only present in the portion of the full-length video of a pair, that portion of the full-length video is identified as comprising overlaid text, and if text is present in the portion of the full-length video and the matched textless clip of a pair, that portion of the full-length video is identified as comprising background text;
• the method further comprising training a classifier for the full-length video in dependence on whether each matched pair is identified as being associated with overlaid text or background text.

A device for processing a video file and a textless video file comprising a plurality of textless video clips, wherein the textless video clips are derived from the video file, the device comprising:

• a comparison module for comparing each textless video clip to a plurality of portions of the video file, the plurality of portions corresponding to the full length of the video file;
• a comparison module for determining each textless video clip having an image being similar to only an image of only one portion of the video file as being a matched pair;
• a comparison module for identifying, for each image matched pair, if their text content is different, wherein identification of a different text content dictates that the textless video clip corresponds to a portion of the video file having overlaid text;
• and
• a training module for training a classifier to predict whether an area of text detected in the full length video is overlaid text in dependence on this identification;
• a processing module configured to:
• determine the probability of each portion of the full-length video having overlaid text via the trained classifier;
• determine each textless video clip being similar to more than one portion of the video file as being a potential matched pair;
• and
• resolve the potential matched pairs with the determined probability.

A device for identifying portions of a video which comprise overlaid text, the device comprising:

• an input for receiving the video;
• an input for receiving textless clips/video;
• a comparison module configured to compare each textless clip to portions of the video and identify those textless clips which match only one portion of the video;
• a processor for processing, based on said identification, the textless clips and the matched portions of the video, the processor being configured to:
• detect the text images in each, and
• compare the text images in each matched pair;
• identify those matched pairs as each having a text box as background text;
• identify these matched pairs in which only one has a text box as being overlaid text;
• train a model in dependence on said two identifications,
• receive the video;
• detect portions of the video having text;
• apply the model to said portions to compute a probability that the portion having text has overlaid text;
• thereby providing a probability of overlaid text for each portion having text, in a matching step:
• the processor being further configured to:
• receive the video;
• receive the textless clips;
• identify those textless clips of the video which match more than one portion of the video;
• select only one portion for that textless clip based on the portion having the highest probability;
• wherein all textless clips for which a match is found are uniquely matched to one portion of the video,
• wherein the processor is further configured to:
• receive the video;
• receive textless clips;
• compare each textless clip to portions of the video.

A device for automatically identifying portions of a video which comprises overlaid text, the device comprising:

• a first input for receiving the video;
• a second input for receiving textless clips;
• a comparison module for comparing each textless clip to portions of the video;
• a processor configured to identify those textless clips which match only one portion of the video;
• matching those textless clips which match only one portion of the video to that portion;
• identify those textless clips which match more than one portion of the video;
• selecting one portion of the video for that textless clip based on the portion of the video determined to have the highest probability of containing overlaid text;
• match that textless clip to the selected portion of the video;
• wherein all textless clips are matched to one portion of the video.

A device for automatically identifying portions of a full-length video as being associated with overlaid text, the device comprising:

• an input for receiving the full-length video;
• an input for receiving textless clips of the full-length video;
• an input for receiving a model for identifying portions of the full-length video as comprising overlaid text;
• a processor configured to:
• identify text in portions of the received full-length video, and in dependence on the model apportioning a probability value to such portion, the probability value indicating the probability of the text being overlaid text;
• identify the textless clips which match any portion of the full-length video to provide matched pairs;
• if a textless clip matches only one portion of the full-length video, match that textless clip to that portion of the full-length video;
• if a textless clip matches multiple portions of the full-length video, match that textless clip to the one of the portions having the highest probability;
• and
• identify each portion of the full-length video uniquely matched to a textless clip as containing overlaid text.

A device for training a classifier, which classifier is used to process a full-length video to identify portions of the video which contain overlaid text, the device comprising:

• an input for receiving the full-length video;
• an input for receiving textless clips of the full-length video;
• a processor configured to:
• identify textless clips which match to only one portion of the full-length video to provide matched pairs;
• and
• detect in each textless clip and portion of full-length video of each matched pair whether text is present;
• wherein if text is only present in the portion of the full-length video of a pair, that portion of the full-length video is identified as comprising overlaid text, and if text is present in the portion
• of the full-length video and the matched textless clip of a pair, that portion of the full-length video is identified as comprising background text;
• train a classifier for the full-length video in dependence on whether each matched pair is identified as being associated with overlaid text or background text.