【作者】 杨全；

【导师】 王民；

【作者基本信息】 西安建筑科技大学 ， 计算机应用技术， 2007， 硕士

【摘要】 手语识别是聋人与健全人自然交流的途径，是手势识别的重要研究内容。随着多模式人机接口技术的发展，手语识别的研究逐渐成为人们研究的热点。手语识别可以分为基于视觉(图像)的识别系统和基于数据手套(佩戴式设备)的识别系统。因为基于视觉的手语识别方法交互方式自然、更能反映机器模拟人类视觉的功能，所以目前是手语识别的研究重点。本文采用基于视觉的方法对静态手指语字母手势图像的识别算法进行了研究。手指语字母的识别过程可分为四个部分：图像采集、预处理、特征提取和识别。在预处理部分，对经过标准化处理的手语图像进行灰度变换、图像平滑、Pyramid分割、二值化和边缘提取。在特征提取和识别部分，本文提出了一种在Euclidean距离空间内计算图像几何矩中由7Hu不变矩特征量得到的组合矩特征量和图像间最小距离相结合的融合算法。首先，将经过预处理的手指语字母图像的边缘检测图像反色处理后进行Euclidean距离变换(EDT)，计算其7Hu不变矩特征量和组合矩特征量；然后，对二值图像进行EDT，通过计算待识别图像与模板EDT图之间的最小距离得到图像的一组特征值。最后，通过设定两个特征的权重来计算图像与模板间的距离，对30个手指语字母进行识别，最高识别率为90％。更多还原

【Abstract】 Sign Language Recognition is a communication way between the deaf and the health. As a key part of the hand gesture recognition, it has been paid more attention than before with the development of Human Computer Interaction. There are two ways of it: one is vision-based recognition, another is digital glove-based recognition. The former could imitate the human vision and more natural than the latter, so a vision-based method is used here to do the recognition research.Sign language recognition is composed of four parts: collecting images, preprocessing images, extracting image features and recognition. During the preprocessing, image gray transforming, image smoothing, Pyramid segmentation, single threshold segmentation and edge detection are performed on standard sign language images.In the part of feature extraction and recognition, this paper presents a method based on geometric moment features: A sign language recognition algorithm based on 7Hu invariant moments. The edge image of sign language letter is treated by counter-color, and carried on Euclidean Distance Transform (EDT). Its 7Hu moments and the combination moments are calculated. Then binary image also implements EDT, calculates the minimum distance between the image and the template. Finally, the recognition of 30 letter gestures is performed by computing distance, in which different coefficients are set to these two features. The best recognition rate of 90% is achieved.更多还原