基于MEB-SVM的静态手势识别研究

Static gesture recognition research with MEB-SVM

为了实现快速高准确率的静态手势识别,提出了一种新的支持向量机分类方法。图像分割是图像识别的基本步骤,采用彩色直方图为特征可以将手部区域从背景中分离出来;然后将得到的手部图像去除手部以外的区域,并二值化,得到手的轮廓;再用8邻域方法处理轮廓图,得到一个表示轮廓的坐标序列,再将复数序列采用傅里叶变换,从而得到一个大小、平移、旋转不变的一维矢量。最后对矢量用MBE-SVM进行分类、识别。通过这种方法完成的手势识别正确率达到了90%以上,基本实现了静态手势识别的目标。

Aiming at realizing high speed and high accuracy hand gesture recognition,a new support vector machine（abbreviated as SVM） classifier was proposed.Image segmentation is a first step of image recognition color-histogram method was applied to dig hand area from background;then binaries the hand image got from last step,and then,get the hand gesture＇s contour line;after that 8-connected neighborhood method was used to deal with the contour line to get a serial coordinates denote the main information of hand gesture,then use Fourier transform to get a suitable vector.Finally MEB-SVM was used to classify these vectors and get the meaning of hand gestures.The result indicates that the MEB-SVM method can get an accuracy of 90% in hand gesture recognition.And it can understand most hand gestures meaning.