Secure Image Processing Based On Homomorphic Encryption

In the era of rapid development of information,people have higher requirements for information security.By uploading personal information to the server,the server can safely process the signal according to the user's needs without the participation of a third party.Encryption as a powerful method to protect information from malicious attacks and to ensure the confidentiality of it.With the development of encryption technology,signal processing under encryption has been drawing a lot of people's attentions.The personal information was uploaded to the server by the user,then,to meet the user's needs,the server processes the signal safely.The signal processing in encryption domain starts from the security of the signal.After a series of transformation operations,the signal is applied to the recognition and detection,such as embedding and extraction of watermark signal,recognition and retrieval of image signal,etc.Considering the computational complexity of encryption,this dissertation chooses homomorphic encryption algorithm as encryption algorithm.Homomorphic encryption algorithm can calculate the data when the data is encrypted.Considering the computational complexity and memory space,Paillier algorithm with additive homomorphism is chosen as the security encryption method.Combining the encryption algorithm with the image processing scheme,then,the image signal security processing is realized.The main contents of this paper are as follows:(1)In this dissertation,the nearest threshold method is used to realize the comparison of numerical values in the encryption domain,which overcomes the difficulty that the homomorphic encryption operation does not keep the order.On this basis,this dissertation proposes a flag bit method to reconstruct the signal by recording the size of the comparison number.In this dissertation,a local binary feature pattern of uniform pattern is introduced,and the extracted features are combined with the gray histogram to apply to the recognition scheme.(2)In this dissertation,the framework of adaptive lifting wavelet is introduced,and the image signal is decomposed and reconstructed based on adaptive lifting wavelet transform,which is successfully applied to watermarking and face recognition.In this dissertation,the quantization expansion of adaptive lifting wavelet is combined with the encryption algorithm to realize the distortion free adaptive lifting wavelet transform in the encryption domain.In this dissertation,the local feature extraction scheme is transformed into encryption domain,and combined with adaptive lifting wavelet transform,it is applied to face recognition and watermarking experiments.Experiments show that the scheme of adaptive wavelet transform in encrypted domain is feasible and applicable.(3)From the point of view of linearity and reducing computational complexity,this dissertation proposes a fast wavelet transform based on the existing wavelet transform schemes.The convolution and down sampling of traditional wavelet transform are transformed into the multiplication operation of bit matrix,which reduces the operation time.Fast wavelet transform is used to decompose and reconstruct the image signal.Considering that there may be decimals in the fast wavelet transform operation in plaintext domain,this dissertation solves the problem that floating-point number can not participate in encryption operation by fast wavelet quantization.By combining fast wavelet transform with encryption algorithm,fast wavelet transform in encryption domain is realized.Fast wavelet transform in encrypted domain also uses local Binarization in encrypted domain for feature extraction and face recognition.The security watermark experiment and face recognition experiment verify the applicability of fast wavelet transform in encrypted domain.(4)In this dissertation,the quantization and computational complexity of two kinds of wavelet transform based on encrypted domain are analyzed and compared.Through the analysis,the adaptive lifting wavelet transform based on encryption has lower computational complexity and smaller quantization factor,while the fast wavelet transform based on encryption domain has less quantization and computational complexity than the adaptive lifting wavelet transform.The above comparison results show that the fast wavelet transform based on homomorphic encryption has better performance.