The Impact of Downsampling Methods on Face Recognition in Electronic Identity Card

Electronic identity cards have limited storage capacity, necessitating the downsizing of images to be stored. Downsampling is a method used to reduce the size of images, but it can result in the loss of essential facial features, impacting face recognition performance. Therefore, the selection of an appropriate downsampling method becomes crucial. In this study, we evaluated and compared the face recognition performance of five different downsampling methods, such as Bicubic Interpolation, Bilinear Interpolation, Lanczos Interpolation, Nearest Neighbour, and Inter-Area using the Asian Face Image Database PF01. We measured the face recognition performance using False Rejection Rate (FRR) at various levels of False Acceptance Rate (FAR). Nearest Neighbour had consistently demonstrated the lowest performance across various scenarios, making it unsuitable for downsampling. In contrast, Bicubic Interpolation has consistently outperformed other methods and is favored for downsampling. In cases where downsizing to a much lower size is required, Lanczos Interpolation offers a preferable option. Our experimental results revealed that the choice of the downsampling method significantly influenced face recognition performance up to 8.41% at specific FAR values. This study highlights the critical importance of selecting the right downsampling method to preserve essential facial features, ensuring optimal face recognition performance for electronic identity cards.