Gaussian Mixture Model Mapping in face recognition

It is difficult to appropriately measure the similarity between human faces under different settings, e.g. pose, illumination, expression and shield. In this paper, a new method called Gaussian Mixture Model Mapping (G3M) is proposed to solve the difficulties. The distribution of faces is divided into many Gaussian functions to cover different settings. A generic identity data set, in which each identity contains multiple images with large intra-personal variation, is adopted to construct the Gaussian mixture model. When considering two faces under significantly different settings, we can judge their feature space distribution by Gaussian mixture model and normalize them into standard space. And then, the normalized faces can be compared by feature in standard space. Finally, we use Multi-pie database to compute the spline functions and test this mode, and LFW is also considered. This method can substantially improve the performance in our test experiment.