Thomas Guthier, Volker Willert, Julian Eggert, "Topological Sparse Learning of Dynamic Form Patterns", Neural Computation, vol. 27, no. 1, pp. 42-73, 2005.

Motion is a crucial source of information for a variety of tasks in social interactions. The process of how humans recognize complex articulated movements such as gestures or face expressions remains largely unclear. There is an ongoing discussion on how explicit low-level motion information, such as optical flow, is involved in the recognition process. We contribute to the discussion by introducing a computational model that classifies the spatial configuration of gradient and optical flow patterns. The patterns are learned with an unsupervised learning algorithm based on translation invariant non-negative sparse coding called VNMF that extracts prototypical, generative optical flow patterns shaped e.g. as moving heads or limb parts. A key element of the proposed system is a lateral inhibition term that suppresses activations of competing patterns in the learning process, leading to few dominant and topological sparse activations. We analyse the classification performs of the gradient and optical flow patterns on three real world human action recognition and one face expression recognition dataset. The results indicate that the recognition of human actions can be achieved by gradient patterns alone, but adding optical flow patterns increases the classification performance. The combined patterns outperform other biological inspired models and are competi- tive with computer vision approaches.