Krawtchouk矩在脱机手写汉字识别中的应用

提出一种基于Krawtchouk矩的脱机手写汉字识别方法。与Zernike矩和Legendre矩等连续正交矩特征相比,Krawtchouk矩是数字域的离散正交矩,不存在数字化过程中所带来的近似误差问题,在计算过程中不需要进行坐标转换,而且构造简单,更加适合用来描述数字图像。在此将Krawtchouk矩用于手写大写金额的识别,并在隐马尔可夫模型(HMMs)框架下对其性能进行了测试。实验结果表明,Krawtchouk矩比传统的连续矩更适合用来描述数字图像,识别效果比连续矩有较显著提高。此外,还对不同参数下的Krawtchouk矩性能进行考察。

Using Krawtchouk Moments for Off-line Handwritten Chinese Character Recognition

This paper proposes an approach using the set of Krawtchouk moments and HMMs (Hidden Markov Models) for character recognition.Compared with continuous orthogonal moments such as Zernike and Legendre moments,the implementation of Krawtchouk moments doesn't involve any numerical approximation and coordinates transform,since their basis set is orthogonal in the discrete domain of the image coordinates space.In this paper,Krawtchouk moment features are extracted to recognize off-line handwritten legal amount.Because of the perfect performance for the time-varying signal,HMMs framework is selected as the classifier.Experiments confirm the performance of the new proposed methods.With little increasing of computational cost,the Krawtchouk moments can achieve higher average recognition rate than Legendre and Zernike moments under HMMs framework.Moreover,different parameters for the total recognition system are also studied.