Head Tilt Classification Using Fft-Pca/Svm Algorithm

ABSTRACT

The complexity of a face’s components originates from the constant variations in the facial component that occur with respect to time. Notwithstanding these variations, humans recognize a person very easily using physical characteristics such as face, voice, gait, etc. Automatic face recognition deals with extracting these essential features from an image, placing them into a suitable representation and performing some kind of recognition on them. This study presents an assessment of the performance of Fast Fourier Transform and Principal Component Analysis with Support Vector Machines (FFT-PCA/SVM) under the constraint of head tilting. 80 head pose images from 10 individuals were extracted from the Massachusetts Institute of Technology Database (2003-2005). Classification rate and runtime were adopted as the numerical evaluation methods to assess the performance of the study algorithm. All computations were done using MATLAB. An earlier study asserted that the higher the degrees of head-pose the larger the Euclidean distance and that above 20°, the Euclidean distances become profoundly larger compared to the 4° head-pose. The study therefore divided the data set into two classes; Lower (below 20°) and Upper (20° and above) and performed an SVM classification. Results from the classification showed that 16° head tilts were more likely to be in the upper class than the lower class. This indicates that with the classification, the Euclidean distance becomes larger from 16° and beyond. It was observed that the SVM classifier performed better in the upper class (20° and above) than that of the lower class (below 20°). It is therefore recommended that further studies be conducted to understand the poor performance of the model in the lower class (below 20°).