Learning Streamed Attention Network from Descriptor Images for Cross-Resolution 3D Face Recognition

In this article, we propose a hybrid framework for cross-resolution 3D face recognition which utilizes Streamed Attention Network (SAN) that combines handcrafted features with Convolutional Neural Networks (CNNs). It consists of two main stages: first, process the depth images to extract low-level surface descriptors and derive corresponding Descriptor Images (DIs), represented as four-channel images. To build DIs, variation Local Binary Pattern (3DLBP) operator encodes differences using sigmoid function. Then, design CNN learns from these DIs. The peculiarity our solution in processing each channel input image separately, fusing contribution by means both self- cross-attention mechanisms. This strategy showed advantages over direct application Deep-CNN face; on one hand, DIs can reduce diversity between high- low-resolution data encoding properties are robust resolution differences. On other, it allows better exploitation richer information provided features, resulting improved recognition. We evaluated proposed architecture challenging cross-dataset, scenario. aim, first train network scanner-resolution data. Next, utilize pre-trained feature extractor data, where output last fully connected layer is used descriptor. Other than standard benchmarks, also perform experiments newly collected dataset paired faces. use high-resolution gallery, while faces probe, allowing us assess real gap existing types Extensive benchmarks show promising results respect state-of-the-art methods.