Image Hashing Algorithms Based On Rotation Invariant Uniform Local Binary Patterns And Zernike Moment

Cloud-based multimedia systems are attracting more users due to the benefits of high-end computation infrastructure.However,transmitting multimedia data to cloud servers managed by the third party server raises some security and privacy issues and brings crisis to the credibility of image content.Thus,image authentication has attracted serious concerns in the community of image content security.To address this issue,the image feature-based schemes and image hashing-based schemes are widely used in cloud-based multimedia systems for security applications such as image authentication and tamper detection.The image hashing-based schemes is a one-way expression of content,the effective features of image content are converted into a fixed-sized sequences according a certain rule.Generally,image hashing schemes should meet the robustness and discrimination capability.The robustness means that the image hashing schemes should have the ability to resist the content-preserving operations.And the discrimination capability requires the ability to distinguish the image hashing sequences of two images have different visual content.In addition,the other security applications such as image content authentication is sensitive to content changes,the key is required to control the security of image hashing.In this paper,the main object is the digital image and image hashing schemes is studied.The main work and results are summarized as follows:In order to improve the performance of discrimination capability to some image hashing schemes using single image component,a perceptual image hashing approach is proposed by exploiting rotation invariant uniform local binary patternsLBP_P,R^riu2,NMF and color features,which effectively robust against various content-preserving image operations and shows outstanding classification performance.Specifically,input image is firstly preprocessed to build a normalized image.Next,LBP_P,R^riu2is exploited to build a stable textural feature matrix of image content,and non-negative matrix factorization?NMF?is used for data reduction to provide a shorter length of image hash and construct theLBP_P,R^riu2-NMF feature.In addition,the color feature is generated from the mean value of each color block.Finally,theLBP_P,R^riu2-NMF feature and the color feature are concatenated to form the image hash,which is referred to LBP_P,R^riu2-NMF-Color hashing.Experimental results show that the proposed approach achieves better robustness,including image scaling and small-angle rotation,and discrimination capability than some existing works.To enhance robustness to any-angle rotation and the ability with tampered image authentication,a perceptual image hashing approach is proposed by exploiting Zernike moment and local image entropy,which effectively robust any-angle rotation and capability with tampered image authentication.First,log polar transformation is applied to image;secondly,the Zernike moments of the whole image is calculated,the original image hash is combined to estimate the rotation angle of the received image,and the rotation angle is used to certain the starting position of the local image entropy;then,the bilateral interpolation technique is used to recover the received image according rotation angle and the size of original image,so the local image entropy of the first Q sample image blocks in the Canny edge outline salient points of the restored image is extracted;finally,all features are concatenated to generate the final image hash.Experimental results show that the proposed approach is robust to any-angle rotation and has excellent classification performance in robustness and uniqueness,which is superior to other existing works.Up to present,the image hashing technology still lacks the ability to robustness image geometric attacks and authentication of tampered images.Therefore,it is hoped that the design ideas and research results proposed in this paper will promote the research of image hash schemes.