<![CDATA[Face shape is an important factor in choosing a suitable hairstyle. However, many people find it difficult to recognize their own face shape, so their hairstyle choices often do not suit them. This study develops a Python desktop application that classifies face shape into five categories (Oval, Round, Square, Heart, and Oblong) and displays knowledge-based hairstyle recommendations. Facial landmarks are extracted using MediaPipe Face Mesh. The system computes four geometric measurements (forehead width, cheekbone width, jaw width, and face length) and then uses a hierarchical, rule-based classifier to determine face shape. We tested the system on 60 frontal facial images of students, with ground-truth labels determined by a majority vote among three independent annotators. The class distribution comprised 16 Square, 15 Oblong, 14 Round, 12 Oval, and 3 Heart images. The evaluation yielded an overall accuracy of 78.33% (47/60; 95% Wilson confidence interval: 66.4–86.9%) and a macro-averaged F1-score of 78.5%. The Heart class contains only three images, so its metrics are indicative only, and users or experts have not yet validated the hairstyle recommendations. These results indicate that landmark-based geometric classification is a feasible, lightweight, offline prototype for face shape analysis]]>
Copyrights © 2026
