基于仿射传播聚类的自适应手写字符识别

Adaptive handwritten character recognition based on affinity propagation clustering

对于手写字符识别过程中相似字符较多且相同字符存在大量不规则书写变形的问题,提出一种改进的仿射传播聚类算法加入手写字符识别过程中。该算法基于原始仿射传播(AP)聚类算法,将其与聚类评判函数Silhouette结合,通过AP算法迭代过程自适应地改变偏向参数以调整类别数,并且结合每次聚类质量得到最优聚类结果。基于手写汉字识别的实验结果表明,加入了原始AP算法的识别率比传统识别过程得到的识别率总体提高1.52%,而加入改进AP算法的识别率又比加入原始AP算法的识别率总体提高了1.28%。该实验结果验证了加入聚类算法于手写字符识别过程的有效性,而改进AP算法相比原始AP算法在收敛性和聚类质量上都有一定的提高。

For too many similar words and lots of irregular writing ways of the same words in the handwritten character recognition, a modified Affinity Propagation （AP） clustering algorithm was proposed to add to the recognition process. Clustering judging function Silhouette was combined with original AP algorithm in the proposed algorithm. Class number was updated by changing preference parameter adaptively through iterative process of AP algorithm. And then the optimal clustering result was obtained by assessing clustering quality of every iteration. The experiment of handwritten Chinese character recognition indicates that the recognition rate of recognition process added original AP algorithm is 1.52% higher than the rate of traditional recognition process. And the recognition rate of recognition process added modified AP algorithm is 1.28% higher than the rate of recognition process added original AP algorithm. The experimental results verify that it is effective to add clustering algorithm to the handwritten character recognition process. And compared with original AP algorithm, convergence and clustering quality of modified AP algorithm are also improved.