Invariant image recognition using triple correlations and neural networks

Triple-correlation-based image representations were previously (Delopoulos, Tirakis, and Kollias, 1994) combined with neural network architectures for deriving an invariant, with respect to translation, rotation and dilation, robust classification scheme. Efficient implementations are described in this paper, which reduce the computational complexity of the method. Hierarchical, multiresolution neural networks are proposed as an effective architecture for achieving this purpose.<<ETX>>