High-Precise and Robust Face-Recognition System Based on Optical Parallel Correlator

Abstract

Facial recognition is applied in a wide range of security systems, and has been studied since the 1970s, with extensive research into and development of digital processing. However, there is only available a 1:1 verification system combined with ID card identification, or an ID‐less system with a small number of images in the database. The number of images that can be stored is limited, and recognition has to be improved to account for photos taken at different angles. Commercially available facial recognition systems for the most part utilize digital computers performing electronic pattern recognition.In contrast, optical analog operations can process two‐dimensional images instantaneously in parallel using a lens‐based Fourier transform function. In the 1960s two methods were proposed, the Vanderlugt correlator and the joint transform correlator (JTC). We present a new scheme using a multi‐channel parallel JTC to make better use of spatial parallelism, through the use of a diffraction‐type multi‐level zone‐plate array to extend a single‐channel JTC. Our project’s objectives were: (i) to design a matched filter which equips the system with high recognition capability at a faster calculation speed by analyzing the spatial frequency of facial image elements, and (ii) to create a four‐channel Vanderlugt correlator with super‐high‐speed (1000 frame/s) optical parallel facial recognition system, robust enough for 1:N identification, for a large database with 4000 images. Automation was also achieved for the entire process via a practical controlling system. The achieved super‐high‐speed facial recognition system based on optical parallelism is faster in its processing time than the JTC optical correlator.