Identifikasi Pola Citra Vena Jari Menggunakan Transfer Learning VGG16

GIFFARI YUSRUL KHOIRURIZAL, Ir. Nazrul Effendy, S.T., M.T., Ph.D., IPM ; Ir. Agus Arif, M.T.

2021 | Skripsi | S1 TEKNIK FISIKA

Keamanan data personal telah menjadi sesuatu yang sangat penting pada era perkembangan teknologi seperti saat ini. Teknologi pengaman yang banyak mendapatkan perhatian oleh kalangan peneliti dan pakar teknologi adalah teknologi berbasis pengenalan biometrik vena jari karena teknologi ini menawarkan tingkat keamanan yang tinggi bagi data personal dan murah dalam hal biaya. Pada penelitian ini, pola-pola citra vena jari seseorang akan dikenali oleh metode Transfer Learning Visual Geometry Group 16 (VGG16). Sebelum pola-pola ini diumpankan ke dalam model VGG16, citra vena jari akan mengalami tahap pre-processing terlebih dahulu dengan menggunakan metode Contrast Limited Adaptive Histogram Equalization (CLAHE). Hasil yang didapatkan dalam penggunaan metode Transfer Learning VGG16 terhadap citra vena jari dari Open Finger-Vein Database Shandong University Machine Learning and Application Homologous Multi-modal Traits (SDUMLA-HMT) adalah tingkat akurasi, Area Under Receiver Operating Curve (AURIC), dan F-1 Score berturut-turut sebesar 98,491% ; 99,047% ; dan 97,170%. Hasil ini menunjukkan bahwa metode Transfer Learning VGG16 memiliki performa klasifikasi yang sangat baik meskipun terjadi kemungkinan ketimpangan data.

Personal data security has become a vey important part in the today's technological developments era. The security technology that is getting a lot of attention by researchers and technology experts is finger-vein biometrics recognition-based technology because it offers a high level of security for personal data and cheap in terms of cost. In this research, image patterns of a person's finger-vein will be recognized by Transfer Learning Visual Geometry Group 16 (VGG16) method. Before these patterns were fed into VGG16 model, finger-vein images would first undergo a pre-processing stage using Contrast Limited Adaptive Histogram Equalization (CLAHE) method. The results obtained in the use of Transfer Learning VGG16 method on finger-vein images from Open Finger-Vein Database of the Shandong University Machine Learning and Application Homologous Multi-modal Traits (SDUMLA-HMT) were accuracy rate, Area Under Receiver Operating Curve (AUROC), and F-1 Score respectively 98.491%, 99.047%, and 97.170%. These results indicate that the Transfer Learning VGG16 method has excellent classification performance even though there is a possibility of data imbalance.

Kata Kunci : Biometrik, vena jari, identifikasi, Transfer Learning, VGG16.