Biologically-Based Interactive Neural Network Models for Visual Attention and Object Recognition

The main focus of this thesis is to develop biologically-based computationalmodels for object recognition. A series of models for attention and objectrecognition were developed in order of increasing functionality and complex-ity. These models are based on information processing in the primate brain,and especially inspired from the theory that visual information processingoccurs along two parallel processing pathways in the primate's visual cortex,the ventral pathway and the dorsal pathway. To capture the true essence ofincremental, constraint satisfaction processing in the visual system, interac-tive neural networks were used for implementing our models. Results fromeye-tracking studies on the relevant visual tasks, as well as our hypothesisregarding information processing in the primate visual system, were imple-mented in the models and tested with simulations. As a rst step, a model based on the ventral pathway was developed torecognize single objects. Through systematic testing, structural and algo-rithmic parameters of this model were ne tuned for performing its taskoptimally. In the second step, the model was extended by considering thedorsal pathway, which enables simulation of visual attention as an emergentphenomenon. The extended model was then investigated for visual searchtasks, where one object is to be identi ed among other objects. In the laststep, we focussed on occluded and overlapped object recognition. The modelwas further advanced on the lines of the presented hypothesis, and simulatedon the tasks of occluded and overlapped object recognition. On the basis of the results and analysis of our simulations we have found thatthe generalization performance of interactive hierarchical networks improveswith the addition of a small amount of Hebbian learning to an otherwise pureerror-driven learning. We also concluded that the size of the receptive eldin our networks is an important parameter for the generalization task anddepends on the object of interest in the image. Our results also show thatnetworks using hard coded feature extraction perform better than the net-works that use Hebbian learning for developing feature detectors. We havesuccessfully demonstrated the emergence of visual attention within an inter-active network and also the role of context in the search task. Simulation