Feature-Based Face Recognition for Vending Machine vs. AMBER Alert

Face recognition systems are an important field in computer vision and are currently used to monitor for dangerous persons and track criminals. A face recognition system uses a database of images and compares another image against those to find a match, if one exists. We implemented an original face recognizer in Java and tested it for recall and accuracy with three image sets. For each facial image, we created a fingerprint of 18 features, such as the RGB values for the eye color, the width and height of the face, various ratios etc. We utilized WEKA machine learning to determine which features are most important and give appropriate weights. We found that the distances between the eyes, nose, and mouth were not useful as they vary little between people. Overall, our method achieved very good results. For scenarios like surveillance which require low false negatives, our accuracy rate was 75% (and 2.3% false negatives). For vending machine authentication where low false positives are needed, we had 49.3% accuracy (and 1.5% false positives). Our system often outperformed WEKA since it uses a more flexible classification rule in the form of a similarity score rather than a binary decision tree.