Rully Soelaiman, Diana Purwitasari, Ariadi Retno Tri Hayati


Face recognition is a difficult task mostly because of the inherent variability of the image formation process ranging from the position/cropping of the face and its environment (distance and illumination) is totally controlled, to those involving little or no control over the background and viewpoint. Moreover, those are allowing for major changes in facial appearance due to factors expression, aging, and accessories such as glasses or changes in hairstyle. A solution has been proposed by considering hybrid classification architectures deal with the benefit of robustness via consensus provided by ensembles of Radial Basis Functions (RBF) networks and categorical classification using decision trees. A specific approach considers an ensemble of RBF Networks through its ability to cope with variability in the image formation. The experiments were carried out on images drawn randomly 50 unique subjects totalling to 500 facial images with rotation ± 50 encoded in greyscale. The faces are then normalized to account for geometrical and illumination changes using information about the eye location. Specifically performance true positive by Ensambles RBF1 (ERBF1) increased on ± 13,86% measures up to RBF while ERBF2 by ± 15,93%. On the contrary the false negative rate decreased by amount of ±5,8% for ERBF1 and somewhat less to ±5,6% for ERBF2. When the connectionist ERBF model is coupled with an Inductive Decision Tree - C4.5 - the performance improves over the case while only the connectionist ERBF module is used.

Abstract in Bahasa Indonesia :

Pengklasifikasian wajah berkaitan dengan variasi data misalnya detil - detil kecil dari wajah atau transformasi saat proses pengambilan citra. Pengklasifikasian wajah dengan metode hibrid menggabungkan pembelajaran berbasis Jaringan Fungsi Basis Radial (JFBR) dan Pohon Keputusan Induktif. JFBR digunakan sebagai metode pembelajaran dalam arsitektur jaringan syaraf tiruan. Untuk meningkatkan kemampuan pengenalan dilakukan pengklasifikasian pada Pohon Keputusan Induktif. Selain menjadi metode penghubung pada pengklasifikasian hibrid, Himpunan JFBR (HJFBR) digunakan untuk penyediaan atribut pada pengklasifikasian Pohon Keputusan Induktif. Uji coba dilakukan pada 50 obyek dengan total ± 500 citra wajah dalam format grayscale. Data dipilih dengan memberi variasi mimik wajah, kemiringan (rotasi) data ± 50 dan juga dipengaruhi oleh pencahayaan di dalam ataupun d iluar ruangan. Rata - rata peningkatan keakurasian positif benar yang diberikan arsitektur HJFBR dibanding JFBR sebesar ±13,86% untuk HJFBR1 dan ±15,93% untuk HJFBR2. Namun menurunkan keakurasian negatif benar sebesar ±5,8% untuk HJFBR1 dan ±5,6% untuk HJFBR2. Penambahan pohon keputusan induktif pada metode hibrid memberikan keuntungan selain tetap dapat meningkatkan keakurasian positif benar juga mampu mengatasi permasalahan sebelumnya tentang penurunan keakurasian negatif benar.

Kata kunci: pengenalan wajah, jaringan fungsi basis radial, pengklasifikasian hibrid, pohon keputusan induktif.


face recognition, radial basis function networks, hybrid classification, inductive decision tree.

Full Text:



The Journal is published by The Institute of Research & Community Outreach - Petra Christian University. It available online supported by Directorate General of Higher Education - Ministry of National Education - Republic of Indonesia.

©All right reserved 2016.Jurnal Informatika, ISSN: 1411-0105


free hit counters
View My Stats