On-Chain Video Copy Detection Based on Swin-Transformer and Deep Hashing

Abstract

In recent years, short videos are spreading faster and become higher quality due to edge-cloud technology. People receive information gradually from graphic to video. At the same time as the number of videos spread rapidly, infringing videos are also flooding the Internet. The wild spread of infringing videos on the Internet has brought serious losses to video websites and original authors. Although video copy detection can solve such problems, the detection results are easy to be tampered with, and the detection results are hardly convincing. Based on this, this paper proposes an open, transparent and verifiable video copy detection method, which uses blockchain technology to ensure the transparency and openness of the results. In addition, this method adopts the combination of on-chain and off-chain methods to automatically perform copyright detection by invoking smart contracts on the chain. This mechanism can securely and immutably store video feature values on the blockchain, ensuring the originality of copyrighted works and the ability to verify detection results. Swin-Transformer and deep hashing are used to obtain video features off the blockchain, which can efficiently match the similarity of existing videos. The method of block comparison can greatly shorten the comparison time, which is 1/50 of the traditional comparison time. Experimental results show that this method is very effective in retrieving similar images and detecting the similarity between original and pirated videos.