Aiming at the related problems of character feature extraction in the original handwritten Chinese characters recognition system, effective features of similar words are learned combining with convolution neural network intelligent, and a new optical image recognition system for handwritten Chinese characters is designed. Efficient training to models is carried out with the rich data resources in handwritten cloud platform. The specific similar subsets are formed according to frequency statistics, and the recognition rates are optimized effectively. Experimental results show that the proposed method can significantly improve recognition rate compared with the support vector machine (SVM) and nearest neighbor classifier method.