Retaining expressions on de-identified faces

Abstract

Abstract-The extensive use of video surveillance along with advances in face recognition has ignited concerns about the privacy of the people identifiable in recorded documents. Prior research into face de-identification algorithms has successfully proposed k-anonymity methods that guarantee to thwart face recognition software. However, there has been little investigation into the preservation of the data utility such as gender and expression in the original images. To address this challenge, a new algorithm based on the Active Appearance Model is proposed here. The main attraction of the approach is that of the preservation of the data utility in terms of facial expression, whilst maintaining privacy protection. The former includes not only the preservation of the expression category (e.g. happy or sad), but also the details of the original expression (e.g. the intensity of a smile and movements of the lips). This is considered to be of significant value in real applications of face deidentification, where the given video contains facial images of the same expression with various degrees of intensity.

Privacy protection performance of De-identified face images with and without background

“…The proposed face de-identification methods form clusters based on Euclidean distance in the AAM feature space. Experimental results from our previous work [6], [9] showed that the reidentification risk is near zero if the attacker uses AAM face representation and matches faces based on Euclidean distance as well. To fully evaluate the privacy protection performance of our proposed methods, further evaluation experiments have been conducted which used various face representation models and distance measures, including Eigenface (PCA) [10], Local Binary Patterns (LBPs) [11], Histogram of Oriented Gradient (HOG) [12] and Local Phase Quantization (LPQ) features [13].…”

De-identification for privacy protection in multimedia content: A survey

“…For both algorithms, there are two main problems: they operate on a closed set I, and the determination of the proper privacy constraint k. In order to produce de-identified images of much better quality and preserve the data utility, the Model-based k-Same algorithms [70] are proposed -one of which is based on Active Appearance Models (AAMs) [72] and another based on the model that is the result of mixtures of identity and non-identity components obtained by factorizing the input images. Modifications to the k-Same Select algorithm, in order to improve the naturalness of the deidentified face images (by retaining face expression) and privacy protection, are proposed in [73,74].…”

Face De-identification with perfect privacy protection

“…Option two performs k-Same-furthest face de-identification without consideration of data utility preservation and then restores the lost utility on the de-identified face generated by kSame-furthest. The second option has been tested by the work presented in [12].…”