Computer Vision in the Classification of Saudi Traditional Fashion: An Analytical Study of Design Characteristics

Abstract

This research addresses a methodological gap in the documentation of Saudi traditional garments by developing and evaluating a classification model based on Computer Vision and Deep Learning (utilizing Microsoft Azure Custom Vision). The study examines three specific scenarios: visually distinct authentic costumes, visually similar authentic costumes, and modern designs inspired by heritage. A dataset of 205 images, labeled by name and geographical region, was utilized, with a data split of 80% for training and 20% for testing. Model performance was evaluated using standard metrics, including overall accuracy, the Confusion Matrix, and class-specific indicators for Precision, Recall, and the F1-Score. The results demonstrated high accuracy -exceeding 90%- for visually distinct costumes. However, experiments on similar garments revealed the limitations of general classification and highlighted the necessity for more granular analysis, such as Object Detection, Semantic Segmentation, and Fine-Grained Visual Categorization (FGVC). Furthermore, tests on modern designs proved the model's capability to quantify visual links between heritage origins and contemporary inspirations. The study recommends expanding the dataset, adopting advanced techniques for multi-component image processing, and integrating expert and consumer evaluations to enhance institutional applicability of the findings.