Research On Face Recognition Based On Privacy Protection

With the rapid development of information technology and multimedia technology,digital images have gradually become the carrier of information exchange.Among them,face recognition plays an important role in the food,clothing,housing and transportation industries as well as in various industries,and from the theoretical research to the practical application of the "blowout period".However,face features are weaker than the iris and fingerprints.Not only can you get face photos easily through the Internet,but you can also use a variety of image processing tools to fake faces.All kinds of information leakage incidents in the recent period also indicate the urgency of users to strengthen personal privacy protection.Faced with this situation,how to protect users'privacy safely and effectively has become a key issue to be solved urgently.At present,research content in the field of face recognition focuses on directly recognizing face images,or encrypting and decrypting face images,but the encryption and decryption operations have the disadvantage of high computational complexity.Aiming at the above problems,this thesis mainly studies the face image scrambling algorithm itself and the face image recognition after scrambling,and proposes the scrambling of the face image first,and then combined with the random forest algorithm to identify the global scrambled person.A face image(SFR-RF,scrambled face recognition of random forest),and a method of classifying and identifying a scrambled face image using a convolutional neural network.The main work and innovations of this paper are as follows:1.The traditional Arnold transform is only suitable for processing two-dimensional space isometric images.For this problem,the Arnold transform method for non-equal length images in two-dimensional space is studied.The Arnold transform is to scramble the face photo in the image space domain,that is,to encrypt the image by changing the position distribution of the original image pixel points.The proposed method improves the Arnold transform by non-affinization according to the two-dimensional spatial reversible standard mapping.The improved transform still has the property of area-preserving and periodicity.Through detailed analysis of the positive and negative transforms,it is more suitable for general scrambling scenarios.2.For the face image recognition after scrambling,this paper combines the random forest algorithm and the image features after face scrambling to propose a global face image scrambling algorithm(SFR-RF)in spatial domain recognition.The feature extraction is performed from the scrambled face image by using the offset random subspace sampling scheme.The decision tree is constructed from the randomly selected features,and then the random forest decision is obtained by combining all decision trees.The algorithm uses the scrambled image pixel value feature to establish a random forest model at each feature point,and the forest model models predict the classification of the test samples.The experimental results show that compared with the single decision tree algorithm,the proposed method still has strong feature extraction ability for the face image after global scrambling,and maintains a high recognition rate.3.Aiming at the problem of weak security of SFR-RF algorithm,this paper proposes a face recognition algorithm based on convolutional neural network and block random parameter Arnold transform.Firstly,the key part of the face image is processed by block processing,then the block image is subjected to Arnold transform scrambling method of random parameters.Finally,the block scrambled picture is input into the deep convolutional neural network model for recognition.From the experimental results,the Arnold random parameter scrambling on the block image effectively reduces the correlation of the ciphertext image,and still maintains a high recognition rate for deep neural network recognition.This paper also uses the chaotic map encryption method for secondary encryption.The results show that the correlation of ciphertext images is further reduced,which not only enhances the protection of face privacy,but also has strong robustness to the image recognition after scrambling transformation.