Object recognition is one of the most important, yet the least understood, aspect of visual perception. The difficulties originate from the variations of objects such as view position, illumination changes, background clutter, occlusion and etc. So, the success of any recognition scheme will depend on its ability to cope with these variations. This thesis presents an object recognition paradigm robust to these variations using local Zernike moments and the probabilistic voting method.
The proposed method using Zernike moments is robust to rotation, scale changes, illumination changes, background clutter and occlusion. The original Zernike moments are normalized by (0,0) Zernike moment which makes the feature robust to scale illumination changes. The modified Zernike moments are calculated around corner points which are extracted from images represented in scale space. For object recognition, we have developed a probabilistic voting method which is an extension of simple voting method. The proposed probabilistic voting method is based on the stability of model Zernike moments and the similarity between the model Zernike moments and the input Zernike moments. This method is better than the simple voting which has high risk of misrecognition for similar objects.
The object recognition system is validated through various experiments. The experimental results show the robustness of the proposed object recognition system.