Face Detection Using the Siamese Network with Reconstruction Supervision

Deep neural networks are important algorithms involved in image classification problems. Among them, Siamese Network is a special type of neural network architecture, whose goal is to calculate the similarity between two given images, which is often applied to distinguish facial similarity. However, how to improve the accuracy of their recognition is still a problem that needs to be solved. To address this issue, this article proposes a new Siamese Network structure, the Reconstructing Supervision Siamese Network. This structure is inspired by the BP neural network and innovatively introduces the idea of BP neural network into the structure of Siamese Network. It has the comprehensive evaluation ability compared to the original graph, further improving the accuracy of the calculated results and improving the classification accuracy. The feasibility and rationality of this method were verified through comparison with traditional Siamese Network. Especially, the added reconstruction supervision process can quickly converge the loss value and demonstrate better accuracy than traditional Siamese Network methods, having an at least 30% improved accuracy. Finally, under the testing of the classic ORL face dataset, this algorithm can be effectively applied to practical face detection and classification problems.