Innovative Techniques for Recognition of Faces Based on Multiresolution Analysis and Morphological Filtering

In this paper, we introduce two new methods for face recognition of frontal images. The methods combine the well known Karhunen-Loeve transform with morphological and subband analysis. The use of this kind of analysis contributes to better discrimination between different images. Morphological and subband approaches are compared while the former is a non linear method and the latter a linear one. The results, obtained using 100 test images, show that both approaches are quite efficient. However, the morphological technique seems to lead to slightly better results (5% , 12% error respectively) while the subband technique has the advantage of decreasing the complexity of the task.