VERIFIKASI TANDA TANGAN MENGGUNAKAN MODEL BIT-M-R50X1 DAN SUPPORT VECTOR MACHINE (SVM)

Safarudin, Yulian (2025) VERIFIKASI TANDA TANGAN MENGGUNAKAN MODEL BIT-M-R50X1 DAN SUPPORT VECTOR MACHINE (SVM). S2 thesis, Universitas Mercu Buana-Menteng.

[img] Text (COVER)
55422120001 - Yulian Safarudin - 01 Cover - YULIAN SAFARUDIN.pdf

Download (426kB)
[img] Text (BAB I)
55422120001 - Yulian Safarudin - 02 Bab 1 - YULIAN SAFARUDIN.pdf
Restricted to Registered users only

Download (147kB)
[img] Text (BAB II)
55422120001 - Yulian Safarudin - 03 Bab 2 - YULIAN SAFARUDIN.pdf
Restricted to Registered users only

Download (455kB)
[img] Text (BAB III)
55422120001 - Yulian Safarudin - 04 Bab 3 - YULIAN SAFARUDIN.pdf
Restricted to Registered users only

Download (362kB)
[img] Text (BAB IV)
55422120001 - Yulian Safarudin - 05 Bab 4 - YULIAN SAFARUDIN.pdf
Restricted to Registered users only

Download (769kB)
[img] Text (BAB V)
55422120001 - Yulian Safarudin - 06 Bab 5 - YULIAN SAFARUDIN.pdf
Restricted to Registered users only

Download (145kB)
[img] Text (DAFTAR PUSTAKA)
55422120001 - Yulian Safarudin - 08 Daftar Pustaka - YULIAN SAFARUDIN.pdf
Restricted to Registered users only

Download (149kB)
[img] Text (LAMPIRAN)
55422120001 - Yulian Safarudin - 09 Lampiran - YULIAN SAFARUDIN.pdf
Restricted to Registered users only

Download (2MB)

Abstract

Verifikasi tanda tangan adalah aspek penting dalam administrasi individu dan lembaga keuangan, terutama untuk mencegah pemalsuan yang dapat menimbulkan dampak hukum serius. Berdasarkan data Direktori Putusan Mahkamah Agung, tercatat 471 kasus pemalsuan tanda tangan selama 2021–2023, sehingga diperlukan metode verifikasi yang akurat. Penelitian ini menggabungkan model BiT-M-R50x1 dan Support Vector Machine (SVM). Dataset yang digunakan dalam pengujian berasal dari Kaggle dengan 2.149 gambar tanda tangan asli dan palsu. Pengujian dilakukan dengan menggunakan model BiT-M-R50x1 dengan preprocessing noise removal, skeletonization, region of interest (ROI), merging of images, ImageDataGenerator dan ekstraksi fitur Grey Level Co-occurrence Matrix (GLCM) dan Red Green Yellow (RGY). Hasil penelitian menunjukkan bahwa preprocessing tambahan seperti noise removal, skeletonization, region of interest (ROI), merging of images, serta ekstraksi fitur GLCM dan RGY menghasilkan performa lebih rendah dibandingkan metode tanpa preprocessing dan ekstraksi fitur. Kombinasi BiT-M-R50x1 dan SVM dengan kernel linear memberikan hasil terbaik pada validation set (accuracy 0,9970; precision 0,9935; recall 1,0000; F1 score 0,9967) dan test set (accuracy, precision, recall, dan F1 score 1,0000), baik dengan maupun tanpa preprocessing ImageDataGenerator. Pengujian model tanpa preprocessing dan ekstraksi fitur pada dataset yang dirusak dengan blur dan noise dengan jumlah kerusakan dataset 25%, 50% dan 75% dari seluruh jumlah dataset menunjukkan penurunan performa, tetapi kernel linear tetap memberikan hasil terbaik di semua tingkat kerusakan. Penelitian ini menyimpulkan bahwa BiT-M-R50x1 dan SVM dengan kernel linear adalah kombinasi optimal untuk verifikasi tanda tangan, sementara preprocessing dan ekstraksi fitur tambahan tidak selalu meningkatkan performa. Signature verification is a crucial aspect in individual and financial institution administration, particularly to prevent forgery that can lead to serious legal consequences. According to data from the Supreme Court Decision Directory, there were 471 cases of signature forgery recorded between 2021 and 2023, highlighting the need for accurate verification methods. This study combines the BiT-M-R50x1 model and Support Vector Machine (SVM). The dataset used in the testing was obtained from Kaggle and consists of 2,149 images of genuine and forged signatures. Testing was conducted using the BiT-M-R50x1 model with preprocessing techniques such as noise removal, skeletonization, region of interest (ROI), image merging, ImageDataGenerator, and feature extraction using Grey Level Co-occurrence Matrix (GLCM) and Red Green Yellow (RGY) color spaces. The research results indicate that additional preprocessing steps such as noise removal, skeletonization, region of interest (ROI), image merging, and feature extraction using GLCM and RGY yielded lower performance compared to methods without preprocessing and feature extraction. The combination of BiT-M- R50x1 and SVM with a linear kernel provided the best results on the validation set (accuracy: 0.9970; precision: 0.9935; recall: 1.0000; F1 score: 0.9967) and the test set (accuracy, precision, recall, and F1 score: 1.0000), both with and without ImageDataGenerator preprocessing. Testing the model without preprocessing and feature extraction on datasets corrupted with blur and noise, with dataset corruption levels of 25%, 50%, and 75% of the total dataset, showed a performance decline. However, the linear kernel still delivered the best results across all levels of corruption. This study concludes that the combination of BiT-M-R50x1 and SVM with a linear kernel is optimal for signature verification, while additional preprocessing and feature extraction do not always enhance performance.

Item Type: Thesis (S2)
NIM/NIDN Creators: 55422120001
Uncontrolled Keywords: Verifikasi tanda tangan, BiT-M-R50x1, Support Vector Machine (SVM) Signature verification, BiT-M-R50x1, Support Vector Machine (SVM)
Subjects: 600 Technology/Teknologi > 620 Engineering and Applied Operations/Ilmu Teknik dan operasi Terapan > 621 Applied Physics/Fisika terapan
Divisions: Pascasarjana > Magister Teknik Elektro
Depositing User: FHADHILAH SHAFA ARISTA
Date Deposited: 13 Feb 2025 06:21
Last Modified: 13 Feb 2025 06:21
URI: http://repository.mercubuana.ac.id/id/eprint/94192

Actions (login required)

View Item View Item