Research On Image Differential Privacy Protection And Visual Security Evaluation Algorithm

In the process of image transmission and processing,there are many security problems,such as image stealing,forgery and so on.Meanwhile,with the development of data mining,machine learning and other technologies,massive image data has been further analyzed and utilized,but it can lead to the disclosure of sensitive information.Image encryption and image privacy protection technologies are important ways to achieve image security protection.The security analysis of the image protection scheme is reflected in the ability to resist cryptanalysis and visual security.Visual security measures the difficulty of unauthorized users in performing visual analysis attacks to obtain valuable information without cracking the key.Most of the existing visual security evaluation algorithms have better evaluation effect when the image quality is good or very poor.Based on the comprehensive consideration of security and processing cost,different applications require different visual security for images.These evaluation algorithms have application limitations.Differential privacy is an effective means of privacy protection,so it has been concerned by scholars at home and abroad and began to apply it to image privacy protection.The existing image differential privacy publishing scheme only guarantees the security of image pixels,and does not evaluate from the perspective of visual security,nor does it verify the relationship between the privacy protection intensity and the degree of image visual disclosure.This paper mainly studies how to accurately evaluate the visual security of images.In addition,this paper also proposes an image differential privacy publishing strategy.The main contributions of this paper are as follows:(1)Build an image visual information calculation model,and based on this model,an evaluation algorithm based on the fusion of edge similarity and information entropy is proposed.According to the chaotic degree of image pixel distribution,different weights are adaptively assigned to the two,and the final visual security index is obtained comprehensively.Finally,the feasibility and robustness of the algorithm are verified on typical encrypted images and video frames.(2)In order to quantitatively evaluate the performance of the visual security evaluation algorithm,the local texture distribution of the image was further extracted by using log-gabor filter and local binary mode(LBP)operator,and the objective evaluation score was obtained by using support vector regression(SVR)to fuse the edge similarity,local texture distribution and local entropy.In order to verify the accuracy,monotonicity,stability and robustness of the evaluation algorithm,low-quality images are selected from multiple image quality databases.(3)An image publishing strategy based on local differential privacy is proposed to ensure that the published image has a strict privacy definition.At the same time,the published image is evaluated from the perspective of visual security,to ensure the consistency of privacy protection intensity and image visual security,and to provide visual security measure for image differential privacy protection.Experimental results show that,compared with the pixelized images,our strategy can effectively reduce the success rate of re-identification attacks.At the same time,the method is very efficient,and it can meet the time-critical requirements.Based on the calculation of the visual information of the image,this paper proposes a broader evaluation algorithm to measure the visual security of images,and provides a new way for the differential privacy protection of the image.