Automatic TV advertisement movie clip detection method based on multi-mode features

Abstract

The invention relates to an automatic TV advertisement movie clip detection method based on multi-mode features. The method includes the steps that a, TV video data are divided into camera lens units; b, an index is made for audio frequency fingerprint features extracted in the step a; c, similar parts of other positions in a video are searched for in the index; d, video matching sections obtained through audio frequency comparison in the step c are in a candidate status; e, matching is refined by searching for boundary points of the matching; f, results generated by multiple video section matching pairs in the step e are subjected to postprocessing, and fragments repeated in time and content are excluded; g, feature recognition is further used for judging whether result video sections in the step f are advertisement movie clips or not; h, after the judgment in the step g is finished, the result is output finally. The automatic TV advertisement movie clip detection method based on the multi-mode features has the advantages that the automatic advertisement movie clip sections are efficiently, accurately and automatically found, the advertisement movie clip sections are automatically found in an unsupervised mode, and the redundant results are excluded through the postprocessing.

Description

technical field [0001] The invention relates to the field of video intelligent analysis, in particular to an automatic detection method for TV commercial clips based on multimodal features, which can realize the combination of repeated segment detection, audio matching and visual matching. Background technique [0002] With the rapid development of the Internet and traditional media, digital media is gradually becoming an indispensable and important information carrier in people's work and life. As an extremely important commercial means, advertising is an important part of digital media applications. With the improvement of people's living standards, TV media, as a way of entertainment, is also constantly enriching. For the wider influence and popularity of their products, merchants are increasingly putting product-related advertisements in TV programs, especially in the middle of prime time and popular programs, where there are endless advertisement clips, but for the majo...

AI Technical Summary

Problems solved by technology

There are two main practical issues: 1) There is a large amount of TV video data, and the data of multiple TV stations 24 hours a day, the practical system implementation requires high efficiency and fast; 2) The accuracy of the detection results of commercial clips, including the accuracy of the boundaries sex

[0005] Technical solution of prior art 1: Learning-based advertisement recognition: collect a batch of advertisement video segments as training data, take the shot unit, and consider the context shot information before and after, extract a series of audio and visual features, and use SVM to train the classifier, Also extract features from the test video data, and identify each shot to determine whether it is an advertisement. It has disadvantages (1) It is necessary to prepare training data in advance, which is inconvenient to use; (2) The recognition calculation is time-consuming and inefficient

[0006] The technical solution of prior art 2: Unsupervised repetitive segment detection in TV video: extract audio and visual features from TV video segments, and combine and use them for clustering at the same time, so as to find similar video segments, which has disadvantages (1) Simultaneous extraction and use of audio and video features failed to make good use of the characteristics of fast audio processing speed and relatively high video accuracy. When processing a large amount of data, the efficiency was not high, and the accuracy of the results was not good; Carry out the refinement search of the boundary points of the advertising film, without post-processing

Images