Face-changing video tampering detection method based on face feature time domain stability and application thereof

Abstract

The invention discloses a face-changing video tampering detection method based on face feature time domain stability and application thereof. The method comprises the steps of decoding a to-be-detected video into a frame sequence and storing the frame sequence frame by frame; detecting a face area of each frame and extracting coordinate values of face feature points; selecting a specific feature point according to the position and activity of the feature points, and calculating a connection vector of the specific paired feature points as a face vector; calculating a deflection angle of a corresponding face vector on an adjacent frame in a spatial position; taking the deflection angle values of all face vectors of two adjacent frames as feature values, and constructing feature vectors; inputting the feature vectors of the training sample into a support vector machine for training to generate a classification model; and classifying to-be-detected video frames by the trained classification model, and judging whether the video is tampered or not. According to the method, detection is carried out by utilizing inherent characteristics of human face biological signals, a relatively good detection effect is achieved, traces of tampered videos are captured by utilizing time domain characteristics of the face vectors, and the method has good universality.

Description

technical field [0001] The invention relates to the technical field of face-changing video tampering detection, in particular to a face-changing video tampering detection method based on the temporal stability of human face features and its application. Background technique [0002] With the improvement of computing power and the explosive growth of data, artificial intelligence has ushered in a new wave of development. In recent years, the technology of using deep networks to generate face-changing videos has become a hot topic. The realistic face-changing effect makes detection more and more difficult. The bigger it is, the easier it is to make face-swapping videos. The abuse of these face-swapping technologies has had a negative impact on society. Malicious tampering not only leads to the infringement of personal portrait rights, but also poses a major threat to social security. Therefore, for The research on face-swapping video tampering detection technology is of great ...

AI Technical Summary

Problems solved by technology

[0002] With the improvement of computing power and the explosive growth of data, artificial intelligence has ushered in a new wave of development. In recent years, the technology of using deep networks to generate face-changing videos has become a hot topic. The realistic face-changing effect makes detection more and more difficult. The bigger it is, the easier it is to make face-swapping videos. The abuse of these face-swapping technologies has had a negative impact on society. Malicious tampering not only leads to the infringement of personal portrait rights, but also poses a major threat to social security. Therefore, for The research on face-swapping video tampering detection technology is of great significance

[0003] In the current technology, most of the detection of face-changing tampering is based on single-frame images. However, since the face-changing video is tampered with by frame-by-frame face-changing operations, the face-changing effect within a single frame has reached a very high level. It brings great difficulty to the detection, so the tampering analysis that only considers a single frame image has limitations, and the correlation of image frames in the time domain is not considered, resulting in poor detection results

In addition, the face-changing video tampering detection technology in the existing technology is popular to use the deep neural network for detection. The accuracy rate of this kind of technology in the library can be as high as more than 90%, which has a good detection effect, but the generalization performance of most algorithms is insufficient. , the accuracy rate drops severely when cross-library testing, and the computational complexity is high, which takes a lot of time

Images