Research On Image Security Based On Fusion Of Compressive Sensing

Multimedia communication plays an important role in our daily life.While it brings the convenience to our daily life,it also brings the problem of security.For example,the sensitive information can be easily stolen,modified,copied and spread without authorization.As a new information sampling theory,the essential idea of compressive sensing is to take random projection to obtain measurements while sampling and compressing the signal.With the received adequate measurements,the signal receiver takes advantage of the sparsity of image to reconstruct the signal by solving the convex problem.Compressive sensing has the advantage of compressing and encrypting the data while sampling.This good property has attracted extensive attentions of image security researchers.In this thesis,the work of the fusion of compressive sensing and image security is analyzed and compared in the aspects of image encryption,image hash,data hiding and secure image retrieval.The state-of-the-art research status and existing problems are analyzed,and some solutions are given,which is beneficial to boost the application of compressive sensing in image security.The existing joint image compression and image encryption schemes have the shortcomings of high complexity and low compression efficiency.This thesis proposes the fusion algorithm of compressive sensing and image encryption.After the image is divided into blocks,discrete wavelet transform is applied to each block and chaotic scrambling is applied to permute the wavelet coefficient.Then chaotic sensing matrix based on structured random matrix is applied to compress and encrypt the permuted coefficient.At last,double random phase encoding and fractional Fourier transform are applied to obtain the cipher image.The experiments prove that the proposed method has the properties of good image recovery performance,high security level,and the resistance to chosen plaintext attack,known plaintext attack and noise attack.To ensure trustworthiness,multimedia authentication techniques have emerged to verify content integrity and prevent forgery.Various image hash schemes have been proposed in literatures for image authentication.The existing image hash schemes do not have good capabilities of discrimination between content-preserving operations and tampering.And if the tampering does not have sparse representation,the image hash schemes have limited capabilities of tampering recovery.This thesis proposes an image hash algorithm to solve this problem by fusing compressive sensing and low rank representation.Firstly,preprocessing is applied to image.Low rank representation is adopted to extract the primary feature and discrete wavelet transform is applied to obtain the feature vector.Then compressive sensing is adopted to encrypt and compress the feature vector.At last,quantization and gray coding are applied to measurements to generate the image hash.At the receive side,the receiver takes the same operations as the sender to extract the feature vector from the received image and applies compressive sensing to compress and encrypt the feature vector.Then he requests the image hash from the authentication server and decodes the image hash to authenticate the received image.At last,compressive sensing reconstruction algorithm is adopted to recover the primary feature and low rank representation is adopted to recover the tampering.Experiments prove that the proposed method has good robustness to content-preserving operations,such as rotation,cropping,additive noise,filter and JPEG compression.Compared with existing works,it has the properties of good discrimination between content-preserving operations and tampering,and performs well in tampering recovery.The joint encryption and data hiding method can protect the content transmitted and verify content integrity.The existing schemes of data hiding in encrypted images have the shortcoming of robustness and cannot be used to transmit images in lossy channel.Some watermarking schemes based on compressive sensing have been proposed which can be used to transmit images in lossy channel but have the shortcoming of large bit error rate of watermark extraction.In this thesis,the scheme of watermarking the encrypted image based on compressive sensing is proposed.In the encryption algorithm based on compressive sensing,the discrete wavelet transform is applied to image to obtain the low-frequent,middle-frequent and high-frequent wavelet band.And each wavelet band is permuted through Arnold map and compressive sensing is applied to each permuted wavelet band to obtain measurements.At last,these measurements are arranged to obtain the cipher image.The scalar costa scheme is adopted to embed watermark into the encrypted domain corresponding to the middle-frequent measurement.The operations of decryption and watermark extraction are commutative.The receiver can extract the watermark and then decrypt the image.He can also decrypt the image directly to obtain the watermarked image and then extract the watermark.The proposed scheme has the properties of high embedding capacity,high correct bit extraction rate and good robustness to cropping attack and noise attack and hierarchical security.In the application of privacy-preserving content-based image retrieval,the essential issue is to construct the image feature protection method and search method effectively and efficiently.The image feature protection method based on homomorphic encryption has the advantage of high security but has the shortcomings of high computation complexity,high communication cost and large cipher image size expansion.The three distance-preserving randomization methods are not cryptographic primitives but random permutation methods.In order to solve the low effectiveness and efficiency problems of the existing works,this thesis proposes the fusion scheme of compressive sensing in the privacy-preserving image retrieval.Firstly,the image feature encryption algorithm based on compressive sensing is proposed.After the image feature is extracted,the wavelet transform is applied to obtain the coefficient and Arnold map is applied to permute the coefficient.Then chaotic compressive sensing is applied to compress and encrypt the permuted coefficient.At last,Arnold map is applied to permute the measurement to obtain the cipher image feature.The generation of image database index and the image database search algorithm are proposed.The experiments prove that the scheme has high security and good tradeoff between effectiveness and efficiency in image search performance.This thesis also proposes the scheme of the privacy-preserving image retrieval based on the encrypted image by fusing compressive sensing.The image encryption algorithm based on compressive sensing is proposed.And then the similarity metric of structure index is adopted to measure the similarity of encrypted images.Experiments prove that compared with existing works,the proposed scheme has the properties of high security level and good search performance.