Hopfield network for affine invariant object recognition

We propose a Hopfield network for affine invariant object recognition. Affine transformation can be considered as an approximation to the weak perspective transformation. Therefore, it is desirable to derive an effective means to establish point correspondence under affine transformation in many applications. By considering the point correspondence as a sub-graph matching problem, a new energy function for affine invariant matching by Hopfield network has been derived. Due to the high order connections, it can only be solved by a fourth order network. However, the order of the network can be reduced by incorporating the neighborhood information available in the data when the proposed method is applied to solve affine invariant shape matching problems. The experimental results show that the proposed method is effective in object recognition under affine transformation