Biologically-Inspired Translation, Scale, and rotation invariant object recognition models

The ventral stream of the human visual system is credited for processing object recognition tasks. There have been a plethora of models that are capable of doing some form of object recognition tasks. However, none are perfect. The complexity of our visual system is so great that models currently available are only able to recognize a small set of objects. This thesis revolves analyzing models that are inspired by biological processing. The biologically inspired models are usually hierarchical, formed after the division of the human visual system. In such a model, each level in the hierarchy performs certain tasks related to the human visual component that it is modeled after. The integration and the interconnectedness of all the levels in the hierarchy mimics a certain behavior of the ventral system that aid in object recognition. Several biologically-inspired models will be analyzed in this thesis. VisNet, a hierarchical model, will be implemented and analyzed in full. VisNet is a neural network model that closely resembles the increasing size of the receptive field in the ventral stream that aid in invariant object recognition. Each layer becomes tolerant to certain changes about the input thus gradually learning about the different transformation of the object. In addition, two other models will be analyzed. The two models are an extension of the “HMAX” model that uses the concept of alternating simple cells and complex cells in the visual cortex to build invariance about the target object