Fast gender classification

This paper presents a fastened algorithm examining the effects of facial features on gender classification. Face images were firstly decomposed by 2-D Discrete Wavelet Transform (DWT2). Different wavelets and different number of filter levels were applied to see the effect of Wavelet Transform on process time and error rate of the method that was proposed in this study. After DWT2, for dimension reduction Principal Component Analysis (PCA) and for gender determination Fisher Linear Discriminant (FLD) were applied to decomposed coefficients. In addition to this, in order to show which facial feature is the most influential for gender classification, parts of several face images, such as, forehead, eyebrows, eyes, nose, lip and chin were masked. Above algorithms were applied to masked face images. Experimental results indicated that the nose is the most influential part for gender classification. Moreover Wavelet Transform decreases process time maintaining the error rate of PCA and FLD. When 1-level DWT2 is used there is no increase in error rate however there is an acceptable increase in error rate when 2-level or 3-level DWT2 is used. 3-level DWT2 decreases process time by 93.4%.