Facial Image Preparation for Virtual Eyeglasses Try-On: A Review

Virtual eyeglasses try-on depends on facial-image preparation, yet the required outputs differ according to whether a system claims visual placement, true-size comparison, support assessment, or scene-consistent appearance. This review examines preparation for 2D overlay, 3D model-based, real-time augmented reality and generative try-on. A hybrid PRISMA 2020-guided protocol combined systematic selection of eyewear-relevant reports with representative selection of generic enabling methods. Of 1952 screened records, 315 reports were included (272 systematic, 43 representative). The evidence is organized into six stages: acquisition and quality assessment, geometric preparation, semantic and occlusion reasoning, existing-eyewear removal, photometric preparation, and personalization-oriented analysis. Recognition-oriented quality scores, normalized landmark errors, scale-aligned reconstruction errors, and whole-face segmentation scores do not establish metric dimensions or regional support accuracy, and eyewear removal infers rather than observes hidden anatomy. No shared benchmark linking facial inputs and preparation outputs to physical frame-fit ground truth was identified. Analytical examples illustrate scale and pose error propagation, and twelve open challenges form a dependency-based roadmap emphasizing calibrated references, regional and optical validation, reproducible baselines, and separate evaluation of dimensional fit, contact, appearance, comfort, and usability.