针对西夏文字识别的特征提取及分类器研究

发布时间：2018-11-19 10:39

【摘要】：字符识别是机器识别领域中的一个传统课题,并且取得了许多研究成果,汉字和古文字的识别是中文信息处理领域的重要研究课题。机器识别研究成果已经进入商业化,广泛地应用于人脸识别、指纹识别、车牌识别、办公自动化和金融商业事务中。字符识别虽有许多困难,由于汉字在实际应用中十分重要,在理论研究方面也有很大意义,然而仍有许多研究坚持不懈的研究这方面的工作。西夏文字的识别目前属于一个待开发的新领域,根据研究表明,基于汉字形体的西夏文字的识别研究存在诸多方面的困难。第一,古文字西夏文有6000余字,因此属于大字符集;第二,西夏文字和汉字相比,其结构更复杂,笔划繁杂,且绝大多数文字笔划数都高于14划,因此西夏文是相似度极高的字符集。第三,手写西夏文字大多都有不同的尺寸和点阵,使得西夏文字的识别更困难、更复杂。古文字数字化最重要的工作是古文字的机器识别,而文字识别中的特征提取是文字识别研究的基础,因此在本文中重点介绍了对西夏文特征提取的算法以及过程。本文首先介绍了西夏文识别的研究意义以及国内外研究现状;然后对西夏文字图像进行预处理,包括归一化、二值化、平滑、细化、倾斜校正处理;然后采用haar-like算法和Gabor小波算法级联方式对西夏文字图像提取特征,最后用AdaBoost算法对提取的特征进行分类识别研究,并将采用单一的haar-like算法和采用haar-like算法与Gabor小波算法级联方式提取特征的分类结果进行比较,取得了较好的分类识别效果。
[Abstract]:Character recognition is a traditional subject in the field of machine recognition, and many research achievements have been made. The recognition of Chinese characters and ancient characters is an important research topic in the field of Chinese information processing. The research results of machine recognition have been commercialized and widely used in face recognition, fingerprint recognition, license plate recognition, office automation and financial and commercial affairs. Although there are many difficulties in character recognition, because Chinese characters are very important in practical application and have great significance in theoretical research, there are still many researches on this aspect. The recognition of Xixia characters belongs to a new field to be developed at present. According to the research, there are many difficulties in the research on the recognition of Xixia characters based on the form of Chinese characters. First, the ancient Xixia language has more than 6000 words, so it belongs to the large character set; Second, compared with Chinese characters, Xixia characters have more complex structure and complicated strokes, and most of them are more than 14 strokes, so the Xixia characters are character sets with high similarity. Third, most of the handwritten Xixia characters have different sizes and lattice, which makes it more difficult and more complex to recognize the Xixia characters. The most important work in the digitization of ancient characters is the machine recognition of ancient characters, and the feature extraction in character recognition is the basis of the study of character recognition. Therefore, this paper mainly introduces the algorithm and process of feature extraction in the Xixia language. This paper first introduces the significance of the research on the recognition of the Xixia language and the current research situation at home and abroad, and then preprocesses the Xixia text image, including normalization, binarization, smoothing, thinning, tilting correction, etc. Then haar-like algorithm and Gabor wavelet algorithm are adopted to extract the features of the Xixia character image. Finally, the AdaBoost algorithm is used to classify and recognize the extracted features. The results of feature extraction using single haar-like algorithm and Gabor wavelet algorithm are compared, and good classification and recognition results are obtained.
【学位授予单位】：宁夏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.43

【参考文献】