Multimedia Copy Detection Technology Based On Robust Hashing

With the rapid development of multimedia and Internet technologies,the production,dissemination and sharing of multimedia files become more and more easy,and they also lead to the number of multimedia files on the network to grow exponentially.The enormous growth of multimedia files not only brings challenges to the retrieval and copyright protection of multimedia files,but also results in the redundancy of data and the waste of network resources.In order to solve these problems,robust hashing was proposed and becomes a hot topic in the field of multimedia processing.Robust hashing maps the features of multimedia into a brief and perceptual digest,and then we can use the perceptual digest for the unique identification of multimedia.Finally,we can determine whether the multimedia file is copied based on the similarity between different robust hashes.As the main carrier of Internet information,images and videos have been widely used because of their vividness.Therefore,this study mainly focuses on the problem of image and video copy detection.In this dissertation,a robust image hashing algorithm based on local and global relations among feature points is proposed for image copy detection.The main contributions of this algorithm are as follows:(1)In this algorithm,the relationship between different feature points,instead of the feature points themselves,is used to generate image hashes,which is more robust to modifications applied to the images.Meanwhile,the local and global relationships represent the image information from different prospective.(2)We employ a graph model for obtaining the relationship between feature points.During the relationship mining,we first connect the points using Delaunay triangulation,and the relationships are encoded in the triangles,which are called the RTs(Relationship Triangles).To avoid the mutual influence between RTs,the RTs are considered as vertices to construct an undirected graph,and an independent set is selected from the graph to produce local relationship,meanwhile,the global relationship is generated by Laplacian matrix of this graph.For video copy detection,this dissertation proposes a robust video hashing algorithm based on spherical torus and non-negative matrix factorization(NMF).The main innovations and contributions of this algorithm are as follows:(1)The use of spherical torus partition can capture the richer spatial and temporal information in the video content,and the establishment of spatiotemporal image will be fused the spatial and temporal information,which will ensure that the extracted hashes have a stronger robustness and discrimination.(2)In the proposed algorithm,we apply NMF to the high-dimensional spatiotemporal image for dimensionality reduction,which not only reduces the computational complexity,but also makes the generated hash more compact.