Human face orientation detection using power spectrum based measurements

This work presents a new approach to human face orientation detection. We start the work with establishing a frequency domain model to analyze the orientation distribution of the oriented patterns in an image, according to the power spectrum of its bi-valued image. This model is derived from the perspective of the rendering rules in classical computer graphics. On the assumptions of the crossed distribution rule and symmetry rule of facial textures, we use the model above to define a series of measurements, called power spectrum based measurements, to quickly determine the orientation of a human face. The experimental results on MIT, ORL and BiolD face databases demonstrate our approach is effective and promising.