Experiments of skin detection in hyperspectral images

Abstract

Skin detection in hyperspectral images has many potential applications to health monitoring and surveillance. In this paper we report on two different approaches that we have followed to tackle with this problem. First, the problem is treated as a classification problem using of active learning strategies to achieve a robust classifier in a short numver of interactions. Second, we approach the problem from the point of view of hyperspectral unmixing, looking for skin endmembers that would allow quick detection in large datasets. We test a new sparse lattice computing based algorithm. We provide experimental results over a dataset of human images in outdoors sunny environment.