Color-based SIFT Descriptor Used in Facial Expression Recognition

The research on the development of children's emotion attracts more and more attention recently. And it is noticed that computer technology will provide researchers with approaches and criterions of judgment and evaluation of children's emotion. Many researchers devote their efforts into study of expression recognition. It is known that the most important part of human facial expression recognition is emotion feature extraction. This paper focus on a novel feature extractor-geometrical based SIFT (scale invariant feature transform) descriptor, which is proven to be affine, scale, rotation, and illumination invariance. Later, color space information and geometrical information are combined together, in order to achieve color invariance, with the high performance in recognition. Color-based SIFT performs better than conventional SIFT with respect to color variations. Experiments which have been carried out support the view that color-based descriptor is prior to gray-based method.