An End-to-End Face Compression and Recognition Framework Based on Entropy Coding Model
Bo Lei, Feng Liang, Haisheng Fu · 2021
Recent years, more and more image/video data were produced, which brings great challenge to data storage and transmission. For face recognition and video surveillance scenario, images/videios need to be compressed and transmitted to intelligent back end for analysis. While general image codecs only extract the feature towards pixel or perceptual similarity, ignoring the Rate-Accuracy performance. In this paper, we proposed an learned end-to-end face compression framework based on entropy coding model, jointly optimize face recognition and image compression performance. Compared with traditional codings, such as JPEG and JPEG2000, better Rate-Accuracy and Rate-Distrotion performance can be achieved by the proposed scheme in LFW(Labeled Faces in the Wild) dataset, especially at low bit rate.