| Title: | 藏文联机手写识别特征提取及分类器设计 |
| Author: | 孟凡强
|
| Issued Date: | 2007-06-01
|
| Degree Grantor: | 中国科学院软件研究所
|
| Place of Degree Grantor: | 软件研究所
|
| Degree Level: | 博士
|
| Keyword: | 动态时间弯折算法
; K均值
; 谱聚类算法
; Gaussian滤波器
|
| Alternative Title: | Feature Extraction and Classifier Design of Tibetan Online Handwriting Recognition
|
| Abstract: | 随着人机交互技术的发展，笔式输入以其方便性和易用性变的越来越流行，适合于个人计算机特别是手持移动设备。因此藏文联机手写识别的研究不仅有理论意义，更具有实用价值，藏文文字的自动识别研究就显得非常迫切了，这就要求我们针对藏文文字字型特点的机器识别进行广泛而深入的研究，找出实际有效的解决办法。
藏文联机手写识别属于模式识别的范畴，本文详细论述了藏文联机手写识别系统设计的整个过程，包括藏文手写样本的收集、预处理过程（归一化，加虚拟笔划，平滑处理，调整点的密度）、特征提取过程以及分类器的设计。根据藏文的字型特点，本文提出了一种基于笔划方向的（8方向）特征提取方法，对于每个手写样本生成8 x 8的采样点，在每个采样点上加上一个 Gaussian 滤波器，这里的 Gaussian滤波器是从Gabor 滤波器的Gaussian 包络求得而来的；利用动态时间弯折算法（DTW）计算两个模式之间的相似度，然后利用两种聚类算法（动态聚类算法K均值算法和谱聚类算法）分别进行了分类器的设计，并给出了实验数据，最后对得出的实验数据进行了分析。 |
| English Abstract: | Along with the development of HCI (human computer interaction), the pen-based input becomes more and more popular because of its convenience and easiness, and this input method is particularly suitable for personal computers, especially mobile equipment. Therefore, the study of Tibetan online handwriting recognition is not only theoretical, but also practical value and automatic identification of Tibetan characters becomes very urgent, which requires us to do extensive and in-depth research as for the characteristics of Tibetan characters in order to find effective solutions.
Tibetan online handwriting recognition belongs to pattern recognition and this paper discusses in detail the whole design process of Tibetan online handwriting recognition, including Tibetan handwriting samples collection, preprocessing (normalization, adding imaginary strokes, smooth processing, adjusting the density of points),feature extraction and classifier design. According to the Tibetan font features, this paper proposes a feature extraction method based on the stroke directions (8 directions) and for each samples, a Gaussion filter is added (the Gaussion filter is from Gabor filter); make use of DTW to compute the similarity of the two patterns and then two different clustering algorithms (dynamic clustering K-means and spectrum clustering algorithm) are used to design pattern classifiers. In the end, the experimental data are gained and analyzed. |
| Language: | 中文
|
| Content Type: | 学位论文
|
| URI: | http://ir.iscas.ac.cn/handle/311060/6590
|
| Appears in Collections: | 中科院软件所
|
| File Name/ File Size |
Content Type |
Version |
Access |
License |
|
|---|
| 10001_200428015029040孟凡强_paper.doc（906KB） | -- | -- | 限制开放 | -- | 联系获取全文 |
|
| Recommended Citation: | |
孟凡强. 藏文联机手写识别特征提取及分类器设计[D]. 软件研究所. 中国科学院软件研究所. 2007-06-01.
|
|
|
