摘要
介绍了一个基于隐马尔科夫模型的、采用模糊分割方式的脱机手写英文单词识别系统。该系统由图像预处理、特征提取、基于HMM的训练和识别四个模块组成。图像预处理中包括二值化、平滑去噪、倾斜校正和参考线提取。然后通过宽度不固定的滑动窗提取特征 ,前两组特征是整体形状和象素分布特征 ,另外又引入了Sobel梯度特征。HMM模型采用嵌入式的Baum Welch算法训练 ,这种训练方式无需分割单词。最后用Viterbi算法识别。对字典中的每个单词 。
Our off-line cursive English word recognition system was based on HMM(Hidden Markov Models) and blur segmentation. It was composed of four modules: image preprocess, feature extraction, HMM-based training and word recognition. The image preprocess consisted of four components: binarization, smoothing and noise removing, incline correction, and reference line detection. Through an unfixed-width sliding window, the features of shape, pixel distribution and Sobel gradient were extracted. HMMs were trained by embedded Baum-Welch algorithm without the segmentation of words. After concatenating letter models to word models for every word in the dictionary, the handwritten English word could be recognized by Viterbi algorithm.
出处
《计算机应用》
CSCD
北大核心
2004年第9期41-43,共3页
journal of Computer Applications
