Primate vision: a single layer perception

Abstract

Visual signals play a significant role in learning, as eyes are actively acquiring new data frames every second and extract relevant information for learning of the newly acquired data in terms of pattern recognition, object identification, and so on. An attempt has been made to link functionality to distinct morphologies of retinal ganglion cells (RGC). Each RGC’s are organized in specific modular connectivity patterns with the photoreceptor cells via bipolar cells. Two distinct morphologies, separately integrated to a single layer network of RGCs, suggest multi-scale feature extraction and identification as one of the functional aspects, depending on the spatial spread of the dendrites of an individual neuron. Apart from texture selectivity, the model also suggests image segmentation as the basic functionality of a single-layered network of RGCs, which might be further feed-forward to successive networks for clustering and classification of visual information. The model shows directional edge selectivity as connectivity specific computation whereas the sensitivity toward fine to coarse edges is specific to the dendritic spread of the connected RGC. Later, the proposed model is incorporated in the hmax model designed by Poggio inspired by Hubel and Wiesel’s functional architecture of the striate cortex that produces some significant results in terms of pattern learning and object recognition.