摘要
为了解决基于尺度不变特征变换的特征包(Bo F-SIFT)算法在描述手势图像特征时对特征点分布情况无法确定的问题,提出了空间金字塔特征包算法提取手势图像特征。该算法通过构造图像金字塔改善了传统的Bo FSIFT算法,生成的描述子能有效表征手势图像的局部特征和全局特征,并能表示图像特征点的分布特性。采用直方图相交核支持向量机进行手势识别。在标准数据库上的测试表明,该算法对于10种手语得到了92.92%的正确识别率,验证了算法的有效性。
A novel algorithm based on the spatial pyramid bag of features is proposed to describe the hand image. It is proposed in order to solve the problem that the distribution of feature points cannot be ascertained when using the hand gesture descriptor based on bag of feature of scale invariant feature transform（ Bo FSIFT）. The capability of the Bo F-SIFT can be improved by generating image spatial pyramid. The descriptor can effectively represent the posture by combining the global features and local features of the gesture image,as well as the distribution character of image feature points. Finally,the hand posture recognition is achieved by using the histogram intersection kernel support vector machine（ SVM）. The experiment on standard database demonstrates the average recognition rate can reach 92. 92 % for 10 kinds of gestures recognition,verifying the efficiency and effectiveness of the proposed algorithm.
出处
《智能系统学报》
CSCD
北大核心
2015年第3期429-435,共7页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金资助项目(61103123)
关键词
手势识别
手势图像
尺度不变特征变换
空间金字塔
特征包
直方图相交核
支持向量机
hand gesture recognition
hand gesture image
scale invariant feature transform（SIFT）
spatial pyramid
bag of features
histogram intersection kernel
support vector machines（SVM）
