A Study Of JPEG Image Enhancement In Encrypted Domain

With the booming and widespread application of cloud computing,more and more data involving personal privacy,commercial interests,and even administrative secrets are stored and processed on the cloud platform.To protect data security,it is usually necessary to encrypt data,but traditional signal processing techniques are difficult to handle encrypted data.In order to solve this dilemma,the academic community have proposed the concept of signal processing in the encrypted domain and has received extensive attention in recent years.This thesis focuses on encrypted-domain image enhancement.The purpose of encryption is to protect the content of image and reduce the information leakage,while more information of image is needed to get better enhancement need.In other words,it is difficult to achieve better enhancement with more safe encryption though the less information leakage,and vice versa.It can be clearly seen that there is a trade-off between security and enhancement effects that should be reasonable choiced.This thesis focuses on DCT-domain JPEG image enhancement,improves a classic JPEG image enhancement algorithm(dctScale)and designs two kinds of encryption and decryption algorithms to implement encrypted-domain JPEG image enhancement.The main work are summarized as follows:(1)The frequency-sensitivity-based JPEG image enhancement algorithm(FS-dctScale)is proposed.A detailed analysis on a classic JPEG image enhancement algorithm("dctScale" for short)is conducted.We found that the dctScale uses the same stretching factor for coefficients of different frequencies in the same subband,which does not consider the sensitivity of human vision to different frequencies.Inspired by the above observations,this thesis improves the dctScale algorithm by adjustting the stretching factor of DCT coefficients according to the perception thresholds of different frequencies,which results in a frequency-sensitivity-based JPEG image enhancement algorithm——FS-dctScale.(2)The Random-Distrubance-based encrypted-domain enhancement algorithms:RD-dctScale and RD-FSdctScale are proposed.In order to implement encrypted-domain JPEG image enhancement,this thesis proposes Random Disturbance-based cryptographic algorithms for dctScale and FSdctScale so that the proposed approaches can process ciphertext.As a result,two encrypted-domain enhancement algorithms based on random disturbance are finally obtained:RD-dctScale and RD-FSdctScale.(3)The Logarithmic-transform-based encrypted-domain enhancement algorithms:Log-dctScale and Log-FSdctScale are proposed.The RD-dctScale and RD-FSdctScale need enhance parameters as auxiliary information to conduct decryption,which causes the cloud platform must allocate additional fields for each enhanced encrypted image to record the enhancement parameters.In order to solve this limitation,this thesis further designs a Log Transform-based encryption algorithm,which makes it no longer necessary to use enhancement parameters as auxiliary decryption information.As a result,two log-based encrypted-domain enhancement algorithms were finally obtained:Log-dctScale and Log-FSdctScale.Extensive experiments are performed to test the performance of the proposed enhancement algorithm FSdctScale,two random-distrubance-based encrypted-domain enhancement algorithms RD-dctScale and RD-FSdctScale,and two Logarithmic-transform encrypted-domain enhancement algorithms Log-dctScale and Log-FSdctScale.The results show that the improved FSdctScale algorithm can obtain better enhancement performance than the original dctScale algorithm.Moreover,compared with their corresponding non-encrypted-domain versions,the four encrypted-domain enhancement algorithms ensure the security of data meanwhile maintaining satisfactory image enhancement performance.