Identifikasi Kompatibilitas Warna Pakaian Berdasarkan Warna Kulit Menggunakan Large Language Model Classifier

Kusumah, Rayhan Almer (2025) Identifikasi Kompatibilitas Warna Pakaian Berdasarkan Warna Kulit Menggunakan Large Language Model Classifier. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211115-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Pemilihan warna pakaian yang sesuai dengan warna kulit merupakan faktor penting dalam meningkatkan penampilan seseorang. Penelitian ini bertujuan untuk mengembangkan sistem rekomendasi warna pakaian berdasarkan atribut warna kulit menggunakan teknik pengolahan gambar dan Large Language Model (LLM) classifier. Dataset yang digunakan dalam penelitian ini berupa gambar model fashion runway yang diambil dari majalah fesyen Vogue. Tahapan penelitian diawali dengan segmentasi gambar menggunakan model Self-Correction Human Parsing, lalu dilakukan ekstraksi warna menggunakan K-Means Clustering. Uji coba klasifikasi warna menggunakan LLM lalu dilakukan untuk menemukan model dan metode yang tepat. Berdasarkan hasil evaluasi, penggunaan model GPT 4.1 menggunakan metode Role-Playing Few-Shot menghasilkan performa terbaik dengan F1 score 82,69%. Berdasarkan hasil evaluasi, model dengan performa terbaik digunakan untuk melakukan klasifikasi warna pada dataset. Kemudian, dilakukan pemetaan matriks co-occurrence dan faktorisasi matriks untuk menganalisis relasi antara tipe warna kulit tertentu dengan tipe warna pakaian dan menghasilkan rekomendasi berdasarkan skor relevansi. Hasil evaluasi rekomendasi berdasarkan fashion expert menghasilkan HR@1 66,67%, HR@2 75,00%, HR@3 91,67% dan NDCG@1 0,83, NDCG@2 0,70, NDCG@3 0,84. Diharapkan, penelitian ini dapat menjadi referensi penggunaan LLM untuk multi-class classification dan pengembangan sistem rekomendasi fashion ke depannya.
======================================================================================================================================
Choosing clothing colors that match one's skin tone is an important factor in enhancing personal appearance. This research aims to develop a clothing color recommendation system based on skin color attributes using image processing techniques and a Large Language Model (LLM) classifier. The dataset used in this study consists of fashion runway model images sourced from Vogue fashion magazines. The research stages begin with image segmentation using the Self-Correction Human Parsing model, followed by color extraction using K-Means Clustering. Color classification trials using LLMs were then conducted to identify the most suitable model and method. Based on evaluation results, the GPT 4.1 model using the Role-Playing Few-Shot method delivered the best performance, achieving an F1 score of 82.69%. The top-performing models were then used to classify colors in the dataset. Based on the results, a co-occurrence matrix mapping and matrix factorization were conducted to analyze the relationship between specific skin tone types and clothing color types, resulting in recommendations based on relevance score. Evaluation of the recommendations by fashion experts yielded HR@1 66.67%, HR@2 75.00%, HR@3 91.67% and NDCG@1 0,83, NDCG@2 0,70, NDCG@3 0,84. This research is expected to serve as a reference for the application of LLMs in multi-class classification and the future development of fashion recommendation systems.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Sistem Rekomendasi Fashion, Fashion Recommendation System, Large Language Model, Prompt Engineering, LLM-Classifier, Image Processing
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Rayhan Almer Kusumah
Date Deposited:	29 Jul 2025 02:49
Last Modified:	29 Jul 2025 02:49
URI:	http://repository.its.ac.id/id/eprint/122468

Actions (login required)

View Item