MODEL KLASIFIKASI DAN IDENTIFIKASI SIDIK JARI DENGAN WAVELET DAN SELF ORGANIZING MAP; THE MODELING OF FINGERPRINT CLASSIFICATION AND IDENTIFICATION USING WAVELET AND SELF ORGANIZING MAP

SUWARNO, Sri, Subanar

2015 | Tesis | FMIPA

Classification and identification of fingerprints are two important processes in utilizing fingerprints as a forensics tool. In the classification, fingerprints are grouped into classes based their similarity to narrow the searching process while, in the identification, a fingerprint is matched to all fingerprints in a database to find the highest similarity. There have been some methods of classification and identification proposed in the literature that gave very good results, especially for fingerprints in good condition and had core and delta features. However, it is not easy to detect core or delta in a fingerprint. In this dissertation, a new model of classification and identification is proposed. The model is based on ridge feature. The ridge is always found in fingerprints in any condition. Fingerprints are modeled as a multiset that has elements of blocks that form a fingerprint. The blocks feature that is the slope of ridges is estimated by utilizing Haar wavelet and Self-Organizing Map network. Haar wavelet is chosen for its simple computation so would give fast execution. The Self Organizing Map network is chosen for its unsupervised training algorithm. The slopes are encoded into eight categories, ranging from 00 to 157.50 with a step of 22.50. Based on the codes of every block, the fingerprint is modeled as a multiset. By using this model, the process of fingerprint classification and verification can be conducted using set operations. The model is validated using 4000 fingerprints published by National Institute of Standard and Technology (NIST). It can be concluded from the validation that this model provides the maximum precision of classification of 52.62% when using blocks of size 8×8 pixels, and 50.62% when using blocks of size 64×64 pixels. For identification, the model can achieve the accuracy up to 67%.

Kata Kunci : N