The Research On Privacy-preserving Image Processing In The Cloud

As an important means of recording life and sharing information,images are the most important unstructured data form in the era of big data.Its application field has penetrated into all aspects of human life and work.However,on the one hand,with the wide use of various image acquisition devices such as mobile phones,tablets,and cameras,the scale of the image is showing an explosive growth trend;on the other hand,the tasks of image processing and analysis(e.g.,feature extraction,object detection,target recognition,and image retrieval)usually contains very complex calculation process.Therefore,large-scale image data has far exceeded the storage and computing power of resource-constrained terminal devices.Cloud computing technology has the characteristics of high reliability,strong versatility,scalability,and ultra-large scale.In recent years,it has been rapidly developed and matured.More and more people,enterprises,and organizations choose to migrate data to the cloud.It also provides a viable solution for storage and managing large-scale image data.However,security issues have always been one of the main factors hindering the promotion and popularity of cloud computing.Storing data in the cloud means exposing data to the third parties that are not completely trusted,thus causing users to lose control of the data and increasing the risk of privacy leakage.Encrypting data can effectively protect data security,but it greatly reduces the availability of data.Therefore,how to ensure the security and usability of data simultaneously in the cloud has become a huge challenge for us,and it is receiving more and more attention and research in academia and industry.In this paper,we deeply analyze the security issues faced by different levels of image processing tasks and carry out research on privacy protection for the key technologies from the following four aspects:image acquisition and recovery,feature extraction,feature analysis,and image retrieval.1.To protect the privacy of image data in the acquisition process,we propose a privacy-preserving image acquisition and recovery scheme.On the basis of acquiring image signals by using compressed sensing technology,we use the secret sharing technology to encrypt image signals,and store large-scale image data in cloud servers.This can effectively reduce the storage and computational overheads of end devices during the image acquisition and encryption processes.At the same time,based on the garbled circuit technology,a new collaborative orthogonal matching pursuit(COMP)protocol is proposed,which allows two cloud servers to perform the image recovery task in a collaborated manner.During the process,the client does not need to interact with the cloud servers,which can minimize the communication overhead in our scheme.2.To protect the privacy of data during the image feature extraction process in mobile sensing scenarios,we propose a lightweight privacy-preserving CNN feature ex-traction framework.Firstly,the image data is stored in encrypted form in the edge servers by secret sharing technology.Then,a series of efficient security computing modules based on secret sharing are designed for the security computing problem between edge servers.On this basis,we propose a series of secure interaction protocols with regard to different network layers of convolutional neural networks,allowing two edge servers to perform CNN feature extraction tasks of ciphertext images in a collaborated manner.Compared with existing schemes,not only the accuracy of the convolutional neural network and the security of the user data can be effectively ensured,but also the overhead and delay of the mobile terminal device can be greatly reduced.3.Aiming at the clustering problem in image feature analysis,we propose a privacy-preserving spectrum analysis scheme.Firstly,the secret sharing technology is used to divide the data into two random parts and send them to two cloud servers re-spectively.Then,a series of efficient security computing protocols are designed for various basic operations in the real number domain.The two cloud servers are allowed to collaboratively perform spectral analysis on the received ciphertext data.The graph structure attributes obtained by the user after decryption can be used to support different data mining applications such as clustering.The scheme provides comprehensive privacy protection for the original data,intermediate messages,and structural attributes such as feature values and feature vectors in the spectrum analysis process.It also effectively reduces the storage and computational overhead of the end devices and the communication overhead with the cloud servers.4.To protect the privacy of data during the process of image retrieval,which is a typical image application,we propose a privacy-preserving image retrieval scheme.The scheme combines techniques such as the vector of locally aggregated descriptors(VLAD),the bilinear projection-based binary codes(BPBC),and the asymmetric scalar product preserving encryption(ASPE).The data owners are allowed to store the image data in ciphertext in the cloud server and the users are allowed to retrieve the images in the encrypted image database.The formal security analysis shows that our scheme can meet the security requirements under the known plaintext attack model.And the experimental results show that our scheme also achieves practical retrieval precision and efficiency.The research on privacy-preserving image processing in the cloud will meet the needs of users for secure data operations in the case of massive image data outsourcing.It also has theoretical and practical significance for promoting the healthy development of cloud computing and its applications.