This report describes face image recognition by 2-dimensional discrete Walsh transform and multi-layer neural networks. Neural network (NN) is one of the powerful tools for pattern recognition. In the previous researches of face image recognition by NN, the gray levels on each pixel of the face image have been used for input data to NN. However, because the face image has usually too many pixels, a variety of approaches have been required to reduce the number of the input data. In this research, 2-dimensional discrete Walsh transform is used for reduction of input data and the recognition is done by multi-layer neural networks. Finally, the validity of our method is varified.

The conventional synthesis procedure of discrete time sparsely interconnected neural networks (DTSINNs) for associative memories may generate the cells with only self-feedback due to the sparsely interconnected structure. Although this problem is solved by increasing the number of interconnections, hardware implementation becomes very difficult. In this letter, we propose the DTSINN system which stores the 2-dimensional discrete Walsh transforms (DWTs) of memory patterns. As each element of DWT involves the information of whole sample data, our system can associate the desired memory patterns, which the conventional DTSINN fails to do.

In this letter we present an electronic circuit based on a neural net to compute the discrete Walsh transform. We show both analytically and by simulation that the circuit is guaranteed to settle into the correct values.

An adaptive LMS filtering system is proposed for computing the Discrete Walsh Transform (DWT). The signal to be transformed serves as the 'desired signal' for the adaptive filter, while a set of periodic Walsh sequences serve as the input signal vector for the adaptive filter. The weights of the adaptive filter provide the DWT. The given approach is more efficient in terms of the required computations and memory locations compared with the direct approach. In contract with existing Fast DWT algorithm, the proposed solution provides more flexibility as far as the signal block length is concerned. In other words, the proposed approach is not restricted to a block length N to be of power 2.