Convolutional Neural Networks for Realtime Multi-Faces Verification with Occlusion

The face is a major component of living creature that becomes a feature to distinguish between one and another living creature and also between living creature with inanimate objects such as statues. In the digital era nowadays, faces are used as objects for identification and verification. But the existence of occlusion in the form of glasses has a potential to influence the verification process carried out by the system. Therefore, a system will be build to perform a real time multi-faces verification processes with occlusion using Convolutional Neural Networks. We propose a "Siamese" architecture with 37 convolutional layer, 10 pooling layer, and 3 fully connected layer. On the training and testing we used Labeled Faces in the Wild (LFW) dataset. The image were taken from 5749 people. We also took 165 images from 11 people for testing with image size 96 × 96 for each images. The verification accuracy achieved for the proposed method is 98%.