Prompt-driven sketch-based 3D face recognition

Recently, sketch-based heterogeneous face recognition has gained considerable attention in intelligent security, criminal investigation, and identity authentication. Recognizing the advantages of 3D face models in accurately representing facial spatial structures and retaining robustness against environmental changes, we investigate a sketch-based 3D face recognition task. In this paper, a prompt-driven sketch-based 3D face recognition method is proposed to explore discriminative identity features from face sketches and 3D face models represented by face depth maps under the guidance of text prompts. Specifically, by employing text prompts related to key facial regions as auxiliary information, a region text–assisted local discriminative feature learning module is designed to effectively explore local facial features critical for recognition. Besides, by introducing modality-related text prompts into feature disentanglement, a modality text–guided identity feature constraint mechanism is presented to facilitate the acquisition of accurate identity features, thus boosting recognition performance. Experiments conducted on the sketch-based 3D face recognition benchmark dataset have confirmed the effectiveness of the proposed method.