Research On Quantitative Privacy Assessment Method Of Objects In Digital Images

In recent years,advanced digital cameras and computer vision technology have been widely deployed in many circumstances.While these technologies provide many conveniences,they also expose individuals’ privacy.Digital image privacy protection has recently received increasing attention from the public.However,the overprotection of image privacy by hiding too much visual information can make images meaningless.Objects in an image usually contain visual information that can potentially reveal private information,thus muti-level privacy protectoin should be considered by filtering out visual information of an image using various image-processing techniques,according to the objects’ different degrees of privacy sensitivity(privacy-level).The analysis of object’s privacy-level is subjective to some degree,and it varies from person to person and from culture to culture.The purpose of this paper is to make the computer simulate the human visual and perceptual thinking to provide objective and quantitative privacy assessment of the visual information contained in the image.Meeting this objective faces two main challenges:(1)determining a method of effectively detecting generic objects in a photo for the extraction of saliency information;and(2)determining a scientific method for assessing the visual private information contained in objects.Firstly,according to the definition of privacy and the study of visual saliency,this paper proposes a quantitative privacy-level assessment method based on the visual characteristics of the objects.And the intelligently quantitative analysis,calculation and evaluation of the image object privacy are realized by the steps of image object detection,object classification and object privacy risk assessment.Then,in the process of object detection,we propose a hierarchical saliency detection method that combines a patch-based saliency detection strategy with an objectness estimation strategy to effectively locate salient objects and obtain the saliency information of each object.The proposed method results in a small set of class-independent locations with high quality and a Mean Average Best Overlap(MABO)score of 0.627 at 1150 locations,which is superior to the score of other saliency detection methods.And in the process of object recognition and classification,this paper focuses on face recognition algorithm,and discusses the effect of image preprocessing techniques on kernel-based face recognition algorithm.And a face recognition algorithm that fuses kernel Fisher discriminant analysis and kernel principal component analysis is proposed.The experimental results show that the proposed method optimizes the performance of the algorithm and improves the face recognition rate.The research on face recognition provides support for the extraction of object categorical specificity in the quantitative privacy assessment method.The object privacy assessment depends on both the visual saliency of the objects and on the specific categories to which the objects belong.We build a computational privacy assessment system to scientifically calculate and rank the privacy risks of objects in an image by creating an improved risk matrix and using the Borda count method.The proposed computational privacy assessment method matches human evaluations to a relatively high degree.The experimental chart and data shown in the paper illustrate the feasibility and applicability of the proposed new method.We have researched the privacy-level of different categories of objects included in the image.Finally,from the point of view of the specific object,this paper takes human face as examples to study the quantitative privacy-level assessment method.Firstly,this paper presents a k-Same face de-identification method based on PCA,the experiments can prove that the algorithm presented we proposed can obtain better performance by comparing the recognition rate and SSIM value under different recognition methods.Then we use size of face as a visual clue,and establish a face privacy-level assessment model by using linear regrassion analysis,and then the relationship between the visual clue and its privacy-level is obtained under different recognition rate requirements.Since this paper uses the recognition rate and the distortion level as two objective variables to build the regression model regarding to the de-identificaiton level,which is used to quantify the privacy-level,thus this quantitative privacy assessment method make the evaluation results more objective.