A method of constructing a dictionary for facial image recognition independent of facial direction

The aspect method that uses mosaic features is a method that can perform high-speed, accurate recognition of humans without requiring special subject motion and which is independent of facial orientation. However, there remains the problem of how to construct the optimal dictionary for use in a knowledge system based on the aspect method. In this paper, for the purpose of applying it to applicant recognition systems based on facial recognition, an optimal dictionary construction method is described for recognizing humans independent of facial orientation. First, a dictionary optimization design plan is presented which uses the statistical characteristics of the distances caused by differences in facial orientation, and those caused by differences among the individuals to be recognized. Next, we describe the results obtained from an optimal dictionary that uses faces without changes in expression to give good matching conditions. The span of facial orientations that can be entered into the dictionary is 30 degrees, and it is shown that an appropriate dimensionality for the mosaic features is 12 x 12. Finally, human recognition experiments are performed with actual images taken at the entrance of a room. The robustness of the approach is investigated in a realistic environment in which the dictionary structural states were derived.