中文摘要：

为解决听力障碍者与无障碍者的信息交流问题,对哑语手势自动识别技术进行研究;提出了一种改进的手势识别算法;首先通过YUV肤色分割、图像差分、连通域检测等算法进行预处理,获取完整的手型区域图像;然后对手型的二值图像进行轮廓检测,采用LBP变换与主成分分析进行特征提取与压缩;最后运用支持向量机的机器学习算法构建分类器,对哑语手势进行分类识别;通过对630张手势图像进行实验,结果表明,提出的算法有效提高了识别率与速度,识别率达到94.22％,速度达到0.29s/幅,可以满足哑语交流的实时性要求.

英文摘要：

In order to solve the issue of information exchange between hearing impaired people and normal people, research the sign language automatic recognition technology. Give an improved gesture recognition algorithm. First, obtain the complete hand--type region of the image, using the image preprocess algorithm such as YUV color segmentation, image differencing, connected domain detection. Then process images through contour detection, have the feature extraction and compression by LBP transform and principal component analysis. Finally, use support vector machine as a training machine learning algorithms to build classifier and finish the classification and identification. To research a total, of 630 gesture images, the experimental results show that the algorithm can improve the recognition rate and speed effectively. And its recognition rate reaches 94.22%, speed reaches 0.29s / piece, meet the requirement of real--time communication for sign language.