Statistical Structure Modeling and Optimal Combined Strategy Based Chinese Components Recognition

Extracting perceptually meaningful components plays an essential role in Chinese character studying process. This paper proposes an improved statistical structure modeling method to pick up all meaningful components in one character. Each stroke is represented by the distribution of the feature points both in model component and input character. The stroke relations are effectively reflected by the statistical dependency. The mutual information among strokes can be calculated to measure the importance of relationships. Considering the local features of components´ difference from the whole character recognition, this paper proposes a method based on local feature to select local components rather than the whole character. At last, we adopt optimal combined strategy to select the best component recognition result. By this method, all the components in one character can be achieved.