Sift algorithm to improve performance of Chinese character classifier using linear decision function

Presents a method to improve the performance of a Chinese character classifier. The method examines the candidates of an existing classifier using a linear decision function to select the most probable one. The algorithm together with a complete scheme has been proposed. The possible improvement of the performance has been estimated based on the experimental inspection of the separability of Chinese characters. The result shows that recognition accuracy can be increased dramatically. This means that the algorithm is practical