High Resolution Remote Sensing Image Encryption And Forensics Based On Zeckendorf Representation

High resolution remote sensing(HRRS)images are important infrastructure,so their security is increasingly concerned.The security lies in confidentiality and integrity,the former guarantees that HRRS images information will not be leaked,the latter guarantees that HRRS images will not be tampered or forged.Generally,confidentiality is provided by image encryption,and integrity can be ensured through image forensics.Encryption is responsible for "pre-prevention",forensics is actually "post-inspection",both of which are inseparable for HRRS image security.There are some unsolved problems in existing image encryption and forensics algorithms,which are embodied in :(a)the lack of secure and practical HRRS image encryption algorithm;(b)the lack of spatial steganography with high embedding rate;(c)the lack of local texture feature perceptual hashing algorithm with high availability.Zeckendorf s Theorem derives a new coding system: Zeckendorf representation,which uses 0,1 represent whether the Fibonacci term is included in Zeckdorf decomposition.Focusing on the "big data",high precision and sensitivity to texture features of HRRS images,this thesis strives to achieve the confidentiality and integrity of HRRS images based on Zeckendorf representation.This thesis proposes two image encryption algorithms.The first is a stream ciphering algorithm using Zeckendorf representation.The "big data" feature determines that HRRS images favors stream ciphering,which emphasizes the randomness of key.The paper proves that the probability of the occurrence of number1 in the middle part of Zeckendorf representation is a constant,and this can be used to generate pseudo-random key stream sequence,and corresponding algorithm is named ZPKG.By contrast,the pseudo-random sequence generated by linear feedback shift register(LSFR)is periodic,that means the key is repeated.Logistic chaotic methods have high sensitivity and poor stability to the initial values.ZPKG makes up for the above deficiencies.Experimental results show that the HRRS images look like noise images after encrypting by ZPKG with indistinguishable hardware consumption with LSFR and Logistic chaotic method.The second is an HRRS scrambling algorithm,named ZRE-PMP,using random encoding of Zeckendorf representation and position mapping using prime numbers,in which the random encoding is used to create a "confusion" effect,and the generation of prime numbers is based on Pisano periods.ZRE-PPM can achieve the desired effect within one iteration and the convergence is fast.It is difficult to find large prime numbers,which improves the security of ZRE-PPM.The fly in the ointment is that it takes too long to implement,which is mainly because the Fibonacci number acquisition and modulus calculation is time-consuming.Image forensics includes proactive forensics and passive forensics.Proactive forensics embeds watermarks and then detect forgery by checking damages on watermarks.HRRS image watermarks should meet the requirements of low distortion,semi-robustness and strong stealthiness.The semi-robustness requires that the watermark should be robust to brightness adjustment,JPEG compression,noise filtering and other content persistence operations,and fragile to malicious tampering.Strong stealthiness means that the watermark position is unknowable.Zeckendorf representations has anti-jamming ability,lightweight tail characteristics and randomness.The anti-jamming characteristic is utilized to encode the watermarks to gain error correction ability(named ZCK-ECC).Combining the spectral features and spatial features,which are invariant with the content persistence operations and geometric transformation,as the original watermark,the semi-robustness can be achieved.Pseudo random numbers are generated to determine the embedding position(named Zck-Mask).The thesis improves LSB spatial domain embedding strategy(named ZCK-LSB)by using lightweight tail characteristics,and thus result in low distortion.The experimental results show that the scheme meets the requirements of imperceptibility recognition with high accuracy and good robustness.In what follows,the perceptual hashing of local texture feature is studied.Local texture features are useful for the detection of copy-move forgery,splice forgery and removal forgery.Thus,two texture feature extraction algorithms based on Zeckendorf representation are proposed.The first is an improved equivalent local binary pattern(LBP)method,named Zck-LBP.Zck-LBP can filter high frequency noise by removing isolated singularities,and it has strong dimension reduction ability.The second is a multiple-operations-combined algorithm: Zck-operators.Zck-operators are adaptable to shape and space relations,and they filter noises by using intersection operators,which only retain the common terms in Fibonacci decomposition,and can automatically screen out the noises.Zck-operators can be used in deep learning,for example,replacing the pooling layer of convolutional neural network.If there is a significant difference between the periphery pixel and the center pixel,the Zck-operators will protrude the difference,highlight the edge.When applying to forgery detection,it is suggested to divide the HRRS image into grids,extract Zck-LBP or Zck-operators value of each grid,calculate the difference of Zck-LBP or Zck-operators value of two grids,and then use the "working point" or "optimization" ideas proposed in the paper to determine the detection threshold.Then,by comparing the difference of Zck-LBP or Zck-operators value of two grids with the detection threshold,and the forgery can be easily located.To enhance the robustness to geometric transformation,the geometric corrections are performed by using SIFT(scale invariant feature transformation).The experimental results show that Zck-LBP and Zck-operators are expert at detecting copy-move forgery,splice forgery and removal forgery,with high recall rate and precision rate.They also show good robustness when dealing with geometric transformation and content persistence operations.