Movement Strategies for Learning in Visual Recognition

In this paper we study the role of movement strategies during learning in object recognition models. We show that a simple model, the RBF, can outperform a more complex hierarchical model, the HMAX, when rotation and scale invariance are provided by the training phase. Moreover, we assess the exploitation of temporal information by the RBF using optic flow. The results show that the RBF model can only exploit the temporal information using optic flow when the training and testing trajectories are the same. This work exemplifies the idea that the complexity of the neural mechanisms in object recognition can be understood not only in the brain but also in the interaction between brain, body and environment.