A deep learning-based method for embedding and extracting transparent watermarks in video images

Abstract

The present invention provides a method for embedding and extracting transparent watermarks of video images based on deep learning, which includes the following steps: acquiring video images of different scenes and different time periods; adding transparent watermarks to the video images, and generating corresponding labels; using data enhancement technology , make a video image transparent watermark data set; train the deep network YOLO v3 model and save the training parameters; use the trained deep network YOLO v3 model to identify the watermark encoding position, category information and category confidence of the watermark encoding graphics to be extracted; integrate the watermark Encode graphics to generate complete watermark information. The invention can increase the redundant information of the video watermark, and can extract the complete information of the watermark when malicious shooting and dissemination under different scenes such as local and different light are realized; Propagation can retain watermark information; use deep learning model to extract watermark information, which is faster and more robust than traditional algorithms.

Description

technical field [0001] The invention relates to the technical field of video image transparent watermark embedding and extraction, in particular to a video image transparent watermark embedding extraction method based on deep learning. Background technique [0002] With the rapid development of computer networks and embedded devices, as well as the enhancement of people's awareness of public safety, surveillance devices can be seen everywhere, and surveillance video can be easily stored, copied, and disseminated, which plays a great role in on-site protection and event recurrence. However, the malicious dissemination of surveillance videos and video images has attracted more and more attention. [0003] Adding extractable watermarks to video images has become an important means to track the source of video image leakage. Video watermarking algorithms are generally divided into three categories. The first category is to embed watermarks in DCT coefficients, the second catego...

AI Technical Summary

Problems solved by technology

Video watermarking algorithms are generally divided into three categories, the first category is to embed watermarks in DCT coefficients, the second category is to embed watermarks in motion vectors, and the third category is to embed watermarks in codewords after entropy encoding, but there are the following Insufficient: The algorithm is not strong in compression resistance. After the video with watermark is re-encoded, the watermark in it will be damaged, which is not conducive to subsequent verification

The watermark redundant information is insufficient. After the watermarked video is broadcast by mobile phones, etc., the watermark information is lost and cannot be extracted.

Images