Face recognition is an active research area because of the importance of its
application areas. Face recognition as almost all digital image processing based systems
begins with elementary processing on the acquired images. The elementary processing is
intended to add more to the abilities of consequent processing steps. The elementary
processing phase is in most of the cases processing the image by some filters. In this study,
an elementary phase of Sobel derivative filters and the derivatives followed by thresholding
by Otsu's method effect on features extracted by moments for face recognition is explored.
The preprocessing step with this setup aims to highlight face local features. The moments
used in the study are Hu and Legendre. Feed forward neural network used as the
classification facility. The results of the study indicated that edges do not form a major
differential component in the values of moments. Consequently, moments could be aided
with feature vectors that focus on edges. Also, the study indicated that Legendre is superior
compared to Hu and the union of Hu and Legendre increases correct recognition
probability.