Practical notes on corruption resistance of Hopfield neural network in Chinese characters pattern recall

Abstract

Digital Hopfield neural networks (DHNN) are well known for its pattern recall capability in noisy circumstances. In this paper, a number of tests are conducted for primarily exploring the recall competency of DHNN in restoring a given set of corrupted Chinese characters. The character patterns are separately corrupted with uniformly distributed random noises, limited translations and rotations. Some author-defined parameters are introduced such as character pattern complexity, similarity, full recall rate, etc. in order to quantify the recall quality. The test findings are tabulated and analyzed accordingly. The results are expected to contribute certain referential data to the prospective research on application of DHNN in Chinese characters recognition with machine vision technologies.