Research On Privacy Protection Of Region Of Interest In Face Images Against Automatic Recognition

With the rapid development of multimedia and computer network technology,the acquisition,storage and sharing of digital images become more and more easy.At the same time,automatic digital image processing and analysis technology has been widely developed and applied,so that people can easily identify,analyze and understand images.However,it also facilitates potential malicious users to attack images,especially for face images shared in social media.Malicious users can use these tools to obtain a variety of personal information.Therefore,face image privacy protection against automatic recognition is of great significance.Based on the above problems,this paper introduces in detail the research background,significance and status quo of face image privacy protection against automatic recognition,and summarizes the related technologies of face image privacy protection against automatic recognition.At the same time,in view of the shortcomings of the existing privacy protection methods against automatic recognition for face images,two privacy protection schemes for shared face images based on region of interest editing are proposed.The main contents of this paper include:1)A privacy protection scheme for face images based on the optimal components sequence and template morphing is proposed.This method takes advantage of the differences between face components in visual quality preservation and privacy protection,and proposes an optimal component sequence based on component differences,which improves the visual quality of privacy protected images.At the same time,the method ultilize facial components morphing technology.Where the face image template library for components morphing is generated by BEGAN(Boundary Equilibrium Generative Adversarial Networks)to avoid invading another person's privacy.Compared with the existing methods,this scheme can effectively protect facial privacy,while better maintaining the visual quality of the image.2)A novel face face privacy protection scheme using CNN(convolutional neural Network)based Region of interest editing is proposed.First of all,a novel face ROI calculation method for face privacy protection based on CNN is put forward.A face image set and a set of CNN networks is selected and utilized to calculate the gradient of each face image in an image dataset,and then an average gradient image is obtained from the image gradients.After that,the Sub-ROIs are determined from the average gradient image,and different Sub-ROI reflects different contribution to the output of the CNN.At the same time,the ROI editing can be applied to any existing full face privacy protection methods such as blurring,pixelization,encryption,etc.Compared with the existing methods,it can achieve the same privacy protection capability with less modification and better image quality compared with the existing methods.The schemes proposed in this paper have found the sensitive area of privacy protection in human face.Under the same privacy protection conditions,the visual quality and utility of the image are improved,which has great application potential in the privacy protection against automatic recognition of face images shared by social media.