Improve Face Verification Rate Using Image Pre-Processing and FaceNet
Siritida Kangwanwatana, Tanasai Sucontphunt · 2022 7th International Conference on Business and Industrial Research (ICBIR) · 2022
Face Verification is mostly used in areas where a false output could end up as a huge mistake. Within the last several years, a lot of improvements have been made, but there is still room for improvement. The quality of the image used to verify are not always the best, with issues such as varying lighting condition, face orientation, etc. In most of the cases where face verification is used, there is usually a limited number of images and people who are not already in the trained database. This research paper presents a method that does not require retraining each time there is a new person not in the database using a pre-trained FaceNet model. Improvement of face verification rate is done in this research paper using image pre-processing on the inputted images, such as using MTCNN to select out the face, face alignment, and brightness adjustment. From testing with Caltech’s Faces 1999 (Front) dataset, our proposed method shows an improvement in accuracy.