The paper argues that face recognition is a crucial component for the next generation of "smart" environments, wearable computers, and ubiquitous computing, because these systems must interact with people in a natural, personalized way without requiring explicit input devices. It reviews the evolution of face‑recognition technology from early AI attempts to modern appearance‑based models, describing the mathematical framework of probabilistic appearance models, dimensionality reduction (PCA, Gabor jets, sparse filters) and discriminative classifiers (Fisher discriminant, SVM). The authors discuss how these techniques overcome the curse of dimensionality and enable real‑time operation on standard PCs. The state‑of‑the‑art algorithms that performed best in the FERET benchmark (MIT, USC, UMD, Rockefeller) are summarized, together with their underlying methods (eigenfaces, elastic bunch‑graph matching, Gabor‑jet flexible templates). Commercial implementations (Visionics/FaceIt, Viisage, Miros) that have adopted these algorithms are listed, illustrating current market adoption. The paper then explores novel application domains: smart environments that adapt to identified users, wearable recognition systems that embed cameras and microphones in clothing, and multimodal fusion of face and speech cues using Bayesian networks. Finally, the authors outline future research challenges—robustness to lighting, pose, aging, and the need for multimodal, low‑power, unobtrusive sensors—to achieve truly ubiquitous, human‑like person identification.
Topics
Face Recognition as the Natural Interaction Modality for Smart Environments
Explains why face (and voice) recognition is uniquely suited to personalize interactions in ubiquitous computing, contrasting it with intrusive biometrics like fingerprints or tokens.
Historical Evolution and Probabilistic Appearance Modeling
Chronicles the shift from early neural‑net and eigenface prototypes to modern probabilistic appearance models, detailing dimensionality reduction (PCA, Gabor jets, sparse filters) and the handling of the curse of dimensionality.
State‑of‑the‑Art Algorithms and FERET Benchmark Results
Summarizes the leading FERET‑tested algorithms (MIT, USC, UMD, Rockefeller), their core techniques (eigenfaces, elastic bunch‑graph, Gabor‑jet templates, discriminative classifiers) and reported performance under varying conditions.
Commercial Deployments Leveraging Academic Advances
Describes real‑world products (Visionics/FaceIt, Viisage, Miros) that embed the academic algorithms into surveillance, ID verification, and transaction systems.
Smart‑Environment and Wearable Applications of Face Recognition
Outlines how identified users enable context‑aware services, facial‑expression analysis, and first‑person wearable systems (e.g., eyeglass cameras, memory aids).
Multimodal Fusion of Face and Speech for Robust Person Identification
Presents a Bayesian network that combines face confidence and speech confidence to produce a final identity decision, illustrating the need for complementary modalities.
Future Challenges: Robustness, Unconstrained Conditions, and Ubiquitous Integration
Identifies open research problems—lighting, pose, aging, low‑power hardware, and seamless integration into everyday objects—to achieve the envisioned fourth‑generation computing.