News splitting method and system

Abstract

The invention provides a news splitting method. The method comprises the following steps: obtaining audio frames and video frames of news to be split; preprocessing the audio frame, obtaining mute time periods, and obtaining a sentence segmentation starting time point and a sentence segmentation ending time point corresponding to each mute time period; preprocessing the video frame, and obtaining anchor feature data; extracting video frames corresponding to the sentence segmentation starting time points and the sentence segmentation ending time points to obtain corresponding sentence segmentation starting images and sentence segmentation ending images; judging whether anchor switching exists between the sentence segmentation starting images and the sentence segmentation ending images or not according to the anchor feature data, and when anchor switching exists, extracting a corresponding video frame according to a corresponding mute time period to obtain a pause paragraph; and obtaining to-be-split points of the pause paragraph, and splitting the news video according to each to-be-split point. According to the method, the pause paragraphs with anchor switching are screened out through the anchor feature data, then the to-be-split points are judged to perform news splitting, the processing speed can be increased, and the accuracy is improved.

Description

technical field [0001] The invention relates to the field of video processing, in particular to a method and system for dismantling news strips. Background technique [0002] News is a major channel for people to receive external information on a daily basis. With the enrichment of media information, there is a large amount of original news data. Automatically dismantling these news programs can greatly save labor costs. The current news stripping technology mainly uses methods such as transition recognition, voice fingerprinting, and face recognition. [0003] The method of dismantling news using voice fingerprints first needs to establish a human voice fingerprint library for news programs, which will inevitably result in a large workload in the early stage, and when there are new anchors, the fingerprint library needs to be constantly refreshed and maintained. Moreover, the voiceprint recognition technology currently on the market is immature. Generally, the accuracy rat...

AI Technical Summary

Problems solved by technology

The accuracy of this method is about 93.3%, but because it needs to calculate whether the faces are evenly distributed in time, it will take a long time to count every face that appears in the video; and because some news programs , there are various combinations of anchor broadcasting methods, such as two-person broadcasting, single-person broadcasting at the same time, the lens of the director station can be zoomed in and out, and non-host faces are filtered by constraining the spatial position information of faces in the entire screen. Generalization ability is not very good

[0006] To sum up, the existing news dismantling methods all require a lot of prior knowledge, and the accuracy rate is not high, and the generalization ability is not strong, so the existing technology needs to be further improved

Images