ABSTRAKSI: Pengenalan ekspresi wajah dari suatu citra merupakan salah satu bidang yang masih sangat menarik perhatian banyak peneliti dalam beberapa dekade belakangan ini. Pada pengenalan ekspresi wajah, ada dua tahap yang harus dilakukan, yaitu pengekstraksian ciri dari suatu citra inputan dan pengklasifikasian citra tersebut ke dalam kelas ekspresi tertentu. Tugas akhir ini mencoba menyisipkan tahapan lain yaitu seleksi fitur dengan menggunakan metode Adaptive Boosting Feature Selection (AdaFs), ke dalam sistem pengenalan ekspresi wajah menggunakan metode Gabor Wavelet, pada tahap ektraksi fitur, dan metode Support Vector Machine (SVM), pada tahap klasifikasinya.
Seleksi fitur oleh AdaFs memilih fitur-fitur tertentu, dari kumpulan fitur Gabor, yang dianggap paling mendiskriminasi data dari kelas ekspresi tertentu dengan data dari kelas ekspresi lainnya, yang tidak menyebabkan performansi klasifikasinya menjadi menurun. Pengujian dilakukan dengan mengimplementasikan metode k-folds cross validation, dan membagi data latih dan data uji ke dalam 3 partisi. Hasil uji coba terhadap semua kombinasi partisi data menunjukkan bahwa proses seleksi fitur dapat mempengaruhi tingkat akurasi pengenalan sistem, baik meningkatkan maupun justru menurunkan performansinya, tergantung dari ketepatan pemilihan jumlah fiturnya. Selain itu, pengadaptasian metode voting SVM multiclass One-Against-All (OAA) dan One-Against-One (OAO) juga mempengaruhi tingkat akurasi sistem. Dari hasil pengujian, akurasi tertinggi didapatkan ketika pengujian dilakukan dengan data set ke-3 dengan klasifier OAA dengan fitur yang dipilih berjumlah 371 fitur, dimana tingkat akurasinya mencapai 95%.Kata Kunci : Pengenalan Ekspresi Wajah, Gabor Wavelet, Seleksi Fitur, Adaptive Boosting Feature Selection, Support Vector Machine, One-Against-All, One-Against-OneABSTRACT: In last decades, Facial Expression Recognition from an image still becomes one of many researchers' interests. In Facial Expression Recognition, there are two main steps that play important role for the purpose of recognition, those are feature extraction, that determines image representation, and classification which classify that image into specific expression class. This TA project tries to add another step by using features selection, with Adaptive Boosting Feature Selection as its method, into facial expression recognition system which is built using Gabor Wavelet, as feature extraction method, and Support Vector Machine, as its classification method.
AdaFs selects significant features, from the pool of Gabor features, considered as unique ones, that can discriminate data of one class from other classes, without decreasing performace of the classifier. Experiments are done by implementing the k-folds cross validation in testing stage and dividing data into 3 partitions. The results of all combinations of data partitions indicate that the feature selection process can affect the accuracy of recognition systems, either increase or actually decrease its performance, depending on the exactness of selecting number of features. In addition, the using of voting methods of multiclass SVM, which are One-Against-All (OAA) and One-Against-One (OAO), also affects the accuracy of the system. From the experiments, the highest accuracy, which reached 95% accuracy level, is obtained when testing is done with the use of the third data set and the use of OAA classifier with the number of selected features is 371.Keyword: Facial Expression Recognition, Gabor Wavelet, Features Selection, Adaptive Boosting Feature Selection, Support Vector Machine, One-Against-All, One-Against-One