Research On Robust Video Hash For Video Copy Detection Based On Geometric Invariant

With the rapid increase of internet online video sites,tens of thousands of videos are being uploaded to those websites and shared every day.A considerable number of these videos are illegal copies or manipulated versions of existing videos,which are distorted or changed in various ways such as change of brightness,text insertion,rotate,resize,and cropping etc.Widespread video copyright infringement and huge storage pressure on the video servers impose strong demand for video copy detection techniques.How to accurately and quickly detect the video copies from huge video database become a challenging and significant problem.Video hashing,also known as video fingerprint,aims to identify a video based on its content by extracting distinguish features form the video sequences.It has recently attracted more and more attention for the video copy detection due to the fact that video hashing does not embed any additional information into the video stream.A good hash function must be robust to non-content transforms.Different videos must have different hash values,which are suitable for fast searching in huge database.However currently most video hashing algorithms for copy detection are not robust against both spatial and temporal transforms at the same time,especially large geometric rotation transform.In this paper,we proposed a three-dimensional transform named spatial-temporal Polar Cosine Transforms(ST-PCT)and introduced a geometrically robust video hashing algorithm for video copy detection.The main work and contribution of this paper are as follows:1)A three-dimensional transform named spatial-temporal Polar Cosine Transforms(ST-PCT)is proposed.ST-PCT transform is an improvement of the conventional hybrid DCT-based video coding,which combines one-dimensional DCT transform and twodimensional Polar Cosine Transforms.It can be used to simultaneously extract the geometric invariant features and spatial-temporal features of video.2)Based on the ST-PCT,we proposed a geometrically robust video hashing algorithm for video copy detection.ST-PCT is used to extract the geometry invariant spatio-temporal features,Principal Component Analysis(PCA)and iterative quantization are used to reduce dimension and produce a compact binary hash code for detecting copy videos.3)The proposed robust video hashing algorithm based on geometric invariant is applied to video copy detection.Experimental results show that our proposed video hash method is highly robust against temporal and geometrical transforms,and achieves high accuracy in detecting video copies which have been subjected to common signal and geometrical modifications.Compared with the state-of-the-art hashing methods,our method is more robust to geometric transforms especially large geometric rotation transform.