A Morphological Neural Network-Based System for Face Detection and Recognition
Mohamed A. Khabou, L.F. Solari · 2006
A human face detection and recognition system is designed to run in Microsoft Windows. A Logitech QuickCam is used to capture a gray scale image. The image is processed by a morphological shared-weight neural network to detect the human faces in the input image. Detected faces are matched against images of known candidates in a database using a simple and fast modified closest neighbor algorithm. If a face matches any of the candidates in the database, the face is labeled accordingly; otherwise it is labeled as "unknown". The system was tested under varying lighting conditions and backgrounds using 4 candidates in the database with 10 images per candidate. The system was able to achieve a 100% face detection rate with no false alarms when the candidates were approximately 24-42 inches from the camera. The system was able to achieve a 95% correct recognition rate on a test set that included only persons in the database and an 87% recognition rate on test set that also included persons not in the database