| The main topic addressed in this thesis is to implement color image pattern recognition based on the lateral inhibition subtraction phenomenon combined with a complex log-polar mapping in multiple spatial frequency channels. It is shown that the individual red, green and blue channels have different recognition performances when put in the context of former work done by Dragan Vidacic. It is observed that the green channel performs better than the other two channels, with the blue channel having the poorest performance. Following the application of a contrast stretching function the object recognition performance is improved in all channels. Multiple spatial frequency filters were designed to simulate the filtering channels that occur in the human visual system. Following these preprocessing steps Dragan Vidacic's methodology is followed in order to determine the benefits that are obtained from the preprocessing steps being investigated. It is shown that performance gains are realized by using such preprocessing steps. |
