Automatic human action recognition is a research topic that has attracted significant attention lately, mainly due to the advancements in sensing technologies and the improvements in computational systems' power. However, complexity in human movements, input devices' noise and person-specific pattern variability impose a series of challenges that still remain to be overcome. In the proposed work, a novel human action recognition method using Microsoft Kinect depth sensing technology is presented for handling the above mentioned issues. Each action is represented as a basis vector and spectral analysis is performed on an affinity matrix of new action feature vectors. Using simple kernel regressors for computing the affinity matrix, complexity is reduced and robust low-dimensional representations are achieved. The proposed scheme loosens action detection accuracy demands, while it can be extended for accommodating multiple modalities, in a dynamic fashion.