关键词: 手写汉字识别, 压缩感知, 稀疏表示, l1范数最小化, 观测矩阵, 信号重构

Abstract:

In view of the low recognition rate and recognition accuracy of handwritten Chinese character in marking system, a handwritten Chinese character recognition algorithm based on compressed sensing theory was proposed. The algorithm first randomly sampled the handwritten Chinese character in the marking system, and got its features. Then, it made sparse representation and minimized its l1 norm to get the sparse solution of the sample. Finally, the obtained sparse coefficient was used to identify the categories of the test samples. This method replaced the traditional method of feature extraction with random sampling of signals, which simplified the realization process of the algorithm. At the same time, with the usage of  training dictionary formed by the existing training samples, the complicated training process was avoided. Study results showed that the recognition rate of the algorithm in handwritten Chinese character database ETL9B reached 99.1%.

Key words: handwritten Chinese character recognition, compressed sensing, sparse re-presentations; , l1-minimization, the observation matrix, signal reconstruction