SUBTITLE DETECTION SYSTEM AND METHOD TO TELEVISION VIDEO
A sub-title detection system and method for television video are provided in which a static region detection is performed, and the results of the static region detection are pruned using a feature based on density of horizontal transition pairs, an intelligent adaptive temporal filtering is performed that uses object-level subtitle information and has suitable properties for integration in a streaming video environment, rectangular bounding boxes are computed around the subtitle using height, width and filling-degree constraints, and finally, subtitle bounding boxes are identified by binary classification based on text stroke alignment features.
1 . A method for determining a subtitle in video, comprising:
detecting a static region in a video to generate a pruned static region map; and
performing adaptive filtering on the pruned static region map to identify a bounding box of a subtitle candidate, wherein the adaptive filtering adjusts the filter coefficients differently for areas within or in the neighborhood of the bounding box than for areas outside or far away from the bounding box; and
computing a set of features of the bounding box of the subtitle candidate to determine the bounding boxes of the subtitle.
2 . The method of claim 1 , wherein performing adaptive filtering further comprises performing adaptive filtering with an adaptive temporal filter and wherein adjusting the filter coefficients further comprises adjusting temporal coefficients of the adaptive temporal filter.
3 . The method of claim 1 that is used for one of motion compensated frame-rate up-conversion, 3D positioning of subtitles in 3D TV and image enhancement operations with specific settings of subtitles.
4 . A system for determining subtitles in video, comprising:
a static region detector that detects a pruned static region in a video and generates a pruned static region map;
an adaptive filtering unit, coupled to the static region detector, that performs adaptive temporal filtering on the pruned static region map to identify a bounding box of a subtitle candidate, wherein the adaptive filtering adjusts the filter coefficients differently for areas within or in the neighborhood of the bounding box than for areas outside or far away from the bounding box; and
a bounding box classifier that computes a set of features of the bounding box of the subtitle candidate to determine the bounding boxes of the subtitle.
5 . The system of claim 4 , wherein the adaptive filtering unit further comprises an adaptive temporal filter unit that adjusts temporal coefficients of the adaptive temporal filter.
6 . The system of claim 4 , wherein the system is used for one of motion compensated frame-rate up-conversion, 3D positioning of subtitles in 3D TV and image enhancement operations with specific settings of subtitles.