Cina Vs Jepang: Identifikasi Keunikan Karakteristik Visual Seni Lukis Tradisional Dengan Metode Semi-Supervised Learning

Ajie, Keysa Anadea Aqiva (2025) Cina Vs Jepang: Identifikasi Keunikan Karakteristik Visual Seni Lukis Tradisional Dengan Metode Semi-Supervised Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211028-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until July 2025. Download (21MB) | Request a copy

Abstract

Penelitian ini bertujuan untuk mengidentifikasi keunikan karakteristik visual seni lukis tradisional Jepang dan Cina melalui pendekatan semi-supervised learning. Metode yang digunakan mencakup deteksi patch, ekstraksi fitur menggunakan ResNet50, reduksi dimensi dengan PCA dan UMAP, serta discriminative clustering berbasis K-Means dengan metrik cosine distance. Hasil clustering digunakan untuk pelatihan model SVM dengan pseudo-label, kemudian dikombinasikan melalui skema hybrid voting menggunakan probabilitas SVM dan skor kedekatan terhadap centroid. Dataset terdiri dari lukisan ukiyo-e dan sumi-e sebagai representasi seni Jepang, serta lukisan tradisional Cina. Hasil pengujian menunjukkan bahwa model hybrid voting dengan filtering patch padding menghasilkan akurasi sebesar 99,98% pada klasifikasi patch dan 83,50% pada klasifikasi gambar utuh. Pendekatan ini juga memungkinkan interpretasi terhadap elemen visual khas yang berkontribusi pada prediksi. Diharapkan, penelitian ini dapat menjadi dasar bagi pengembangan sistem pelestarian dan analisis seni tradisional Asia Timur berbasis pembelajaran mesin.
======================================================================================================================================
This research aims to identify the unique visual characteristics of traditional Japanese and Chinese paintings through a semi-supervised learning approach. The proposed method involves patch detection, feature extraction using ResNet50, dimensionality reduction with PCA and UMAP, and discriminative clustering using K-Means. The clustering results are used to train an SVM classifier with pseudo-labels, and predictions are combined using a hybrid voting scheme that integrates SVM probabilities and proximity scores to cluster centroids. The dataset includes ukiyo-e and sumi-e paintings as representations of Japanese art, and traditional Chinese paintings as comparisons. Experimental results show that the hybrid voting model with patch padding filtering achieves an accuracy of 99.98% on patch-level classification and 83.50% on full-image classification. This approach also enables interpretation of visual elements that contribute significantly to predictions. The findings are expected to support the development of computational methods for analyzing and preserving traditional East Asian art.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Seni Jepang, Klasifikasi SVM, Hybrid Voting, Klasifikasi Berbasis Centroid, Japanese Art, SVM Classification, Hybrid Voting, Centroid Based Classification
Subjects:	N Fine Arts > ND Painting T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Keysa Anadea Aqiva Ajie
Date Deposited:	01 Jul 2025 04:23
Last Modified:	01 Jul 2025 04:23
URI:	http://repository.its.ac.id/id/eprint/119297

Actions (login required)

View Item