Recognition of handwritten Katakana in a frame using moment invariants based on neural network

A method of pattern recognition using a three-layered feedforward neural network is described. Experiments were carried out for handwritten katakana in a frame recognition using the neural network. The problem of scale and translation recognition of handwritten characters using the neural network is described, and the relation of the recognition data set to the recognition rate is examined. The normalization of images using moment invariants is examined. First, translation normalization is achieved by translating the origin to the center of gravity of an image. Secondly, scale normalization is executed. Experiments were carried out in which the number of recognition categories was 5, 10, 20, and 46. Furthermore, experiments were carried out where the sets of recognition categories are changed using the Euclidean distance among them. Recognition rate was increased by using this normalization