New invariant pattern recognition system based on preprocessing and reduced second-order neural network

Proposes a new method for shift, scale, rotation invariant pattern recognition system using a normalization algorithm and a shift invariant neural network of order two with reduced input dimension. The normalization scheme normalizes the scale and rotation of deformed patterns using principal component analysis (PCA). The reduced second-order neural network using the combinations of input pattern pixels and PCA has only (2·N)/5 input nodes, where N is the dimension of the input patterns. Experimental results with four types of aircraft data indicate the superiority of the proposed method to the compared system in terms of both learning speed and recognition rates