Video copy detection method and system

Abstract

The invention is applicable to the technical field of multimedia information processing, and provides a video copy detection method and system thereof. The method comprises the following steps: obtaining an image tag Y in the image training set; carrying out copy attack on each image tag Y to obtain a first image X. The training model of depth feature extraction model is constructed, and the depthfeature extraction model is obtained by training. Obtaining a first fusion feature after fusing the first depth feature and the first DCT coefficient feature of each second image, and storing the first fusion feature as a fusion feature database; Extracting a third key frame of the query video, and fusing the second depth feature and the second DCT coefficient feature to obtain a second fusion feature; Each of the second fusion features is matched with each of the first fusion features to obtain a detection result of the query video. Thus, the invention improves the discrimination and robustness of the video features and improves the accuracy of the video copy detection system.

Description

technical field [0001] The invention relates to the technical field of multimedia information processing, in particular to a video copy detection method and system thereof. Background technique [0002] With the development of science and technology and the widespread popularization of the Internet, the problem of digital video piracy has become increasingly prominent. The content-based video copy detection technology is a method to judge whether the video to be detected constitutes piracy and infringement by comparing the similarity between the video to be detected and the original video. , video infringement includes not only copying the original video, but also zooming, cropping, rotating, blurring, inserting text or graphics and other operations on the video. Content-based video copy detection is crucial to video copyright protection and management effect. [0003] Traditional video copy detection techniques are mainly divided into methods based on global features and m...

AI Technical Summary

Problems solved by technology

Although this invention application uses the method of deep learning to extract the deep features of the video, it directly uses the network model that is pre-trained on the image classification task for feature extraction, ignoring the uniqueness of the copy detection task, and does not carry out the copy detection task. Design and retraining, the extracted depth features are not very discriminative. At the same time, this method densely extracts all frames of the video, and inputs all video frames into the neural network to obtain 4096-dimensional deep features, which requires a large amount of calculation and a burden on the machine. Heavy, this method also only obtains the compact video representation of each second of video, which leads to the problem of rough copy positioning results when performing library video matching

[0006] To sum up, the existing copy video detection technology is obviously inconvenient and defective in actual use, so it is necessary to improve it

Images