A Recurrent Temporal Model for Semantic Levels Categorization Based on Human Visual System

Figure 4

The proposed computational model. In this model, four layers are represented. The S1-layer refers to simple neurons in the v1 region and the edge detection process. The C1-layer points to complex cells in the human visual system; the remarkable points of this layer are the maximization as its main operator and the time appearance in it. The output of the C1-layer is an array of spikes, in which the time is based on the salient of the edges. The S2-layer cells are related to the intermediate visual system features. In this study, the S2-layer includes 110 cells, for ten categorizes. The layer performance evaluation is slightly different at the test and training phase in which the two feedbacks, related to expectation influence, occur on this layer. The C2-layer of the model is performing categorization at different levels and does this by counting the spikes.