Image Analysis Based On Homomorphic Encryption Orthogonal Moments And Wavelet Transform

In the field of secure signal-processing,in order to avoid malicious attacks on some valuable information,it is an effective way to operate the signal processing model on the encrypted information directly.In recent years,more and more attention has been paid to its property of protection of information security.Frorn the perspective of image information analysis,it is necessary to study the implementation scheme of image processing processes such as feature extraction and transform domain processing in the encryption domain.Especially when the signal to be processed is a sensitive image involving privacy,such as biological palm print,face image,patent watermark,etc.,the server needs to ensure the confidentiality and security of the information while meeting the users' processing requirenents.In the method of image feature extraction,the orthogonal molent feature is valued for its invariance of rotation,scale and translation.Because the domain of basis function is continuous,the zero-order approximation method will Iead to the numerical integral error;On the basis of discrete wavelet transform scheme,the enhancement wavelet scheme with adaptive thinking is applied to the feature extraction and classification work,particularly aim at biological images such as palmprint and leaf vein images for its adaptability and simplicity.The methods describe above can be optimized in plaintext domain and implemented in encrypted domain.In this paper,one of orthogonal moment algorithm is optimized to reduce the numerical integral error,and a strategy of encrypted image processing method is proposed by combining the homomorphic encryption algorithm.The main work content includes:(1)In this paper,a new orthogonal polynomial is constructed based on the orthogonal polynomial of the Jacobi-Fourier moments and the relation of integral substitution.The new moment named Nun-uniform sampling Jacobi-Fourier moments.The sampling frequency of the moment increases from low frequency to high frequency,and the amplitude of numerical oscillation of the new polynomial decreases significantly.The above reasons reduce the numerical integral error.The new moment has better performance of image reconstruction and classification.(2)A new method to realize new moments in homomorphic encryption domain is proposed.Combining the previously proposed non-uniformly sampled Jacob-Fourier moments with the additive homomorphism of Pallier system,the new moments in the plaintext domain and the image reconstruction process are implemented in the homomorphic encryption domain.Under the combined action of scale factor and integer operation,the arithmetic involved in the new moment calculation and reconstruction process can be converted and implemented in the encryption domain according to the homomorphic law.This paper verifies the feasibility of the scheme and its applicability in image feature extraction and classification in the encryption domain through experiments.(3)A homomorphic adaptive lifting wavelet scheme and a homomorphic LBP scheme are proposed.In the process of image reconstruction,watermarking and classification,the server carries out wavelet transform and LBP local feature extraction directly without accesses the image plaintext infonmation.The operation parameters of the original adaptive lifting wavelet scheme are simple without decimals.The homomorphic comparison problem is also solved.Compared with the global moment feature extraction,the computation cost of this algorithm is lower,and the experiment also proves its applicability in security watermarking and palmprint recognition.