Privacy-Preserving Facial Recognition Based On Temporal Features

This paper proposes a novel approach for privacy-preserving facial recognition based on the new feature computation technique: Local Binary Pattern from Temporal Planes (LBP-TP) that extracts information from only the XT or YT planes of a video sequence; in contrast to previous work that depend significantly on spatial information within the video frames. To our knowledge, this is the first known facial recognition work that does not rely on the spatial plane, nor that requires processing a facial input. The removal of this spatial reliance therefore withholds the facial appearance information from public view, where only one-dimensional spatial information that varies across time are extracted for recognition. Privacy is thus assured, yet without impeding the facial recognition task which is vital for many security applications such as street surveillance and perimeter access control. Experimental results indicate that the proposed method achieves accuracy of 99.56%, 98.19% and 100% for the recent CASME II, CAS(ME)(2) and Honda/UCSD databases respectively. In addition, a 66% reduction in the number of bytes required for storage and recognition was also observed from these experiments. The outcomes of this research demonstrate that privacy in face recognition can be preserved, without compromising its security (i.e., recognition accuracy) and efficiency. (C) 2020 Published by Elsevier B.V.