AI-Powered Forensic Face Drawing: An Overview of the Research on Methodological, Theoretical, and Real-World Defects in Suspect Identification Systems

Abstract

Artificial intelligence (AI) has transformed forensic facial sketching, introducing advanced deep learning architectures for suspect identification in constrained-data environments. This literature survey systematically analyzes the state-of-the-art in AI-driven forensic facial sketching, identifying critical gaps in methodological, theoretical, and practical dimensions. Methodologically, we highlight the lack of comparative studies across deep learning architectures (e.g., GANs, VAEs, diffusion models), the over-reliance on accuracy as the sole evaluation metric, and insufficient investigation of algorithm robustness against noise and distortions. Theoretically, we identify gaps in understanding how AI models interpret and reconstruct facial features from sparse witness descriptions and the limited research on inference mechanisms with incomplete data. Practically, we note deficiencies in real-world scenario testing, user-centric design for forensic practitioners, and system scalability for operational deployment. By synthesizing existing literature, this survey not only identifies these interconnected gaps but also proposes future research directions to develop more robust, efficient, and forensically applicable AI systems. Our analysis emphasizes the need for standardized benchmarks, comprehensive evaluation protocols, and interdisciplinary collaboration to advance the field.