Face Recognition System Using Feature Extraction Method of 2-D Gabor Wavelet Filter Bank and Distance-Based Similarity Measures

Face recognition is a computer vision task which has a capability to recognize human according to their corresponding face features. Face recognition application in computer system have been developed in recent year including surveillance, human-computer interaction, and multimedia application. Some challenges will possibly be undergone in face recognition research mostly lie on deformation of face condition such as pose, expression, illumination, and accessories use that lead to a noticeable impact on recognition rate. In this research, two-dimensional Gabor wavelet filter bank was implemented to extract face image. Gabor filter has characteristic similar to wavelet that is localized in both time and frequency domain able to extract local feature on face image. Convolution process produces a face representation feature called Gaborface in a terms of magnitude information obtained from sinusoidal complex function. Features are formed as a column matrix concatenation which is called feature vector. At the end of process, one distance-based similarity measure, i.e., Euclidean Distance was implemented to perform a test measurement on data features vector with database features. Testing has been done by giving variation in some aspects, i.e., number of training data, face image deformation such pose and expression, and the variant of filter with variation in scale and orientation. The best system, based on this research, is achieved by
$5 \times 8$
filter bank with accuracy up to 98,67%. The proposed method also shows a very well performance in recognizing the expression changes, but it cannot accommodate properly in pose changes.