Klasifikasi Skintone Menggunakan Convolutional Neural Network (CNN) untuk Mencari Persentase Kecocokan dengan Warna Pakaian

Abstract

.. Choosing clothing colors that complement an individual's skin tone often relies on subjective judgment. This study develops a Convolutional Neural Network (CNN)-based system to classify skin tone and calculate clothing color compatibility percentages. A modified MobileNetV2 model was trained on 2,935 facial images categorized into Light, Medium, and Dark classes. Preprocessing included augmentation, normalization, resizing, and face detection using MediaPipe. Model performance was evaluated on an independent test set obtained from Kaggle and separated from training data before model development. The model achieved 94% test accuracy. Following classification, skin color was analyzed in CIELAB and HSV color spaces. Clothing color compatibility was calculated using a weighted scoring method involving five parameters: Hue, Undertone, Value, Saturation, and LAB Distance scores. Compatibility percentages were determined using predefined color-matching criteria. The system was implemented as a web-based application displaying predicted skin tone and compatibility percentages for predefined clothing colors. These findings demonstrate the potential of combining deep learning and color analysis to support more objective and personalized clothing color recommendations.