Research On Image Hash Algorithm For Copy Detection And Tamper Detection

Image hashing algorithm is an important research topic of image retrieval,image authentication and image copy detection.Essentially,it maps any image into sequence,which is called hashing.In practical application,images are often suffered some content-preserving image processing.Therefore,the same or similar images should have the same or similar hash sequence,namely robustness.Hash sequence of different images should be obviously different,namely distinctiveness.In addition to these two basic properties,image hashing algorithm should be controlled by secret key and the correct hash sequence can't be obtained without the correct key,namely security.In this paper,three image hashing algorithms are proposed by discrete wavelet transform,texture features,bit plane,color opponent component and quadtree decomposition.The research results are summarized as follows.1.An image hashing algorithm based on CS-LBP texture and bit image decompositionTexture features of approximate image of wavelet transform and bit plane features of high-frequency information are combined to generate image hash.Specifically,the input image is regularized to the same size and filtered by Gauss low-pass filter.Then the approximate image and high-frequency information are obtained by wavelet decomposition.Then,texture features and bit plane features are extracted from approximate image and high frequency information respectively.Finally,the hash is generated by combining all the features.The experimental results show that the algorithm has good robustness to common image operations,and achieves good distinction between different images.2.A statistical feature hash based on wavelet decompositionAn image hashing algorithm is proposed by extracting the statistical features of rows and columns of approximate image of wavelet decomposition.Specifically,we first adjust the size of the input image and filtered,then the approximate image is obtained by wavelet decomposition,the statistical features of each row and column are extracted from approximate image,and finally the final hash is obtained by calculating the L2 norm of the row and column statistical features.The experimental results show that this algorithm has good robustness to common image processing operations.In the performance test of copy detection,the algorithm can detect common image copy versions,and this algorithm has good detection performance.3.Perceptual hashing for color image based on color opponent component and quadtree structureColor opponent component and quadtree decomposition are used to extract color features and structure features for generating image hashing.Specifical,the input image is first adjusted to the same size,and then the regularized image is processed by Gauss low-pass filtering,the color opponent component and the brigtness image is obtained from preprocessing image.The color feature is extracted from the color opponent component,and the qiadtree structure feature is extracted from the brightness image,Finally,the hash is generated by combining the color feature and the structure feature.In robustness experiments,the algorithm is robust to gamma correction,JPRG compression,watermarking embedding and other content-preserving image processing operations.Different image can be distinguished well.In tamper detection experiments,the algorithm can detect tampered images and locate tampered areas.