A Study on the Effect of ROI Masks on Face Recognition System Using Digital Recorder

In recent years, biometrics authentication has been widely used to realize high security in airports, office buildings and so on. In public area, surveillance cameras, which are combined with a high efficient coding recorder such as MPEG2 or MPEG4 recorder, are set to take face images of pedestrians and car license plates. However, high compression for efficient recording leads to degradation of details in face and characters on car license plates. Consequently, it is important that region of interest (ROI) is detected correctly and recorded with high quality image. Although there are a number of proposals concerning the detection of face areas and license plates, the relationship between ROI masks and face recognition rate or character recognition rate is not cleared. In this work, we investigate the recognition rates for face authentication with compressed face images under limited storage size. Several kinds of ROI mask are compared, and face recognition rates with respect to ROIs are shown