Research On Anonymization Technique Based Privacy Preserving Method On Facial Image

In recent years,with the rapid development of computer technology and electronic technology,an ever-increasing amount of image and video data is being recorded,stored or processed worldwide.With the application of face recognition technology in the field of identity authentication,the extensive use of video surveillance and mass visual media distribution in modern video sharing,social media and cloud services have ignited concerns about the privacy of people identifiable in the scenes.Therefore,measures need to be taken to ensure a sufficient level of privacy protection to perform secure data sharing while at the same time preventing possible cases of misuse.A common solution to address this problem is face anonymization,or face deidentification,a process that conceals personal identifiers present in image and video data and thus preventing the recovery and misuse of identity information,e.g.,preventing face recognition,while at the same time preserving data utility.Early face deidentification techniques mostly centered around naive approaches,such as pixelation or blurring.However,there is an effective attack for naive approaches,e.g.,parrot attack,and naive approaches achieve a high recognition rate under parrot recognition.Recent research has successfully proposed k-Same solutions to overcome the limitations of naive approaches,which provide k-anonymity privacy protection.However,k-Same solutions still have many shortcomings.The utility of k-Same anonymized data is greatly reduced and cannot be used in face analysis tasks.Therefore,the research on face anonymization becomes more and more important.Details of our main works are as follows.First,since k-Same solutions always force every k original faces to share the same deidentified faces,making the diversity within the set of the original faces lost,it's impossible to track certain individual in a k-Same de-identified video.To address the common drawback of the k-Same solutions,we propose an improved method of neural-network-based face deidentification method based on the k-Same-Net idea.This approach takes the feature difference between the identity and the centroid in each cluster into account for target identity estimation,maintaining the diversity of the de-identified faces.Results of comparative experiments demonstrate that the set of de-identified faces is able to mimic the diversity found in the original videos.The main goal of face deidentification is to promote the anonymity of individual identity information contained in face,but the current trend in this area is to ensure that certain face attributes are still retained for face analysis after data anonymization.Our second contribution is that we introduce the expression parameters to further expand the first work and propose a face anonymization method towards emotion analysis.We encode the information of facial expression characteristic as an input parameter of the generative neural network.The generation process is controlled by the identity parameter and the expression parameter.This approach achieve a more nuanced privacy protection,where identity attribute is anonymous,and facial attribute is preserved.The proposed deidentification method greatly increases its applicability in emotion analysis tasks.Finally,extensive experiments validate the effectiveness of the algorithm.