The Research On Image Hash Technology For Authentication Based On Multiscale Curvelet Transform

Image hash technology maps image data into a short binary sequence that canreflect the major visual features of an image. It can be widely used in contentauthentication, image database retrieval, digital watermarking and other fields. Thispaper focuses on some key problems related to image hash, such as generalframework, authentication principle, performance, and the classification of the hashalgorithms, especially the image hash performance about the robustness, distinction,security, and the new generation of multiscale analysis method——curvelet transform.At present, since the research of applying curvelet transform to image hash algorithmis rare, this paper proposes two image hash algorithms based on curvelet transform.The main contributions are as follows:(1) An image hash based on discrete curvelet transform is proposed. The image isfirstly preprocessed, such as a low pass filter, normalized scale, and then decomposedby fast discrete curvelet transform. The low frequency coefficients contained the mainfeature of images and the two detail layer coefficients contained rich edge informationare selected as the feature vector. And chaotic sequence is used to encrypt theeigenvector. Finally, the image hash sequence is obtained by quantization andcompression. Experimental results show that the algorithm has better robustnesscompared to some other hash methods. It is distinction to different images. The chaossystem enhances the security. The algorithm also has a low complexity.(2) An image hash based on human visual system in curvelet domain is proposed.The image is firstly preprocessed, such as a low pass filter, histogram equation. Andthen the eigenvector is extracted via weighting curvelet coefficients based on humanvisual system in curvelet domain. Chaotic sequence is used to encrypt the eigenvectorto get the middle image hash. Finally, the last image hash is obtained by scramblingthe middle image hash based on random disorder algorithm. Experimental resultsshow that the algorithm has a good robustness performance against most ofcontent-preserving manipulations, and also well discriminative capabilities todifferent images. The algorithm is sensitive for shear replacement and randomdistortion. The dual-key mechanism enhances the security. In addition, the hashsequence is compact.