Face verification using convolutional neural network with partial triplet loss on face wearing glasses

The role of face verification in the field of security and law enforcement has become an important part of daily life. However, the challenges involved in face verification still have an impact of inaccuracy on several factors that affect verification performance. One of them is the use of camouflage causing facial occlusion, such as wearing glasses. This research was conducted to find out the effect of wearing glasses on method performance. Therefore, CNN algorithm with siamese architecture is used to extract features from the face and propose to use the partial triplet loss function to partially minimize the optimization problem on the network. In experiment, our proposed approach achieves the lowest loss of 0.953 in training data, the highest accuracy of 68 for verification on face wearing glasses with partitioning and 73 accuracy for verification on face without wearing glasses with partitioning, both in verification data. These results are better than combination of CNN and transfer learning which only achieves 55. Thus, it can be concluded that our proposed method could handle partial occlusion due to face wearing glasses.