结合肤色分割和ELM算法的静态手势识别

Static gesture recognition algorithm of combinating skin color segmentation with ELM

针对目前复杂背景下手势图像识别率不高、识别困难等问题,基于ELM(extreme learning machine),提出了一种快速手势识别方法。结合RGB与HSV两种颜色空间模型,从复杂背景中去除大部分类肤色的干扰,实现手势分割;采用改进的Hu不变矩以及指尖个数对获取的手势轮廓进行描述;利用ELM进行特征数据分类,从而实现实验所采用手势的识别,其中ELM是在单隐层神经网络的基础上提出来的一种新型前馈神经网络,网络结构比较简单,输入权值和偏差随机给定的。在采用ELM识别的同时又用传统的BP网络进行了识别,结果表明:相对于BP网络,ELM具有较快的学习速度和良好的抗差能力,同时识别率比较高,适合静态手势识别。

Aiming at the problems of low image recognition rate and difficulty in gesture recognition under complicated background, a rapid gesture recognition method based on ELM （extreme learning machine） is proposed. Firstly, by combining RGB and HSV color space model and removing most classification from complex background color interference, gesture segmentation is achieved. Then, the improved geometric moment invariants and the number of fingers are used to describe the obtained gesture contour. Finally the generated characteristics data are classified by using ELM, so as to realize gesture recognition. ELM is a new type of feed forward neural networks which is proposed on the basis of single-hidden layer neural network. ELM＇s network structure is simpler, and the input weights and bias are randomly given. Recognition using ELM and recognition using traditional BP network are experimented respectively. The results shows that ELM is suitable for static gesture recognition, for it has faster learning speed and better resistance, and the recognition rate is higher, compared with BP network.