基于层次化BoF模型和Spectral-HIK过滤的手势识别算法

Hierarchical Bag-of-Features with Spectral-HIK filter based hand posture recognition

为了解决基于计算机视觉的人类手势识别问题,提出一种名为层次化Bag-of-Features(BoF)的模型.该模型通过对人手区域进行划分和对图像特征分别向水平和垂直轴投影来提取图像特征的空间分布信息.为了准确快速地实现手势识别,构建一种基于直方图交叉核的手势识别分类算法.该算法结构简单、运行效率高,而且充分利用层次化BoF模型的结构特点.为了进一步提高在复杂背景下手势识别准确率和运行效率,采用一种基于谱和直方图交叉核的背景特征点过滤算法.实验结果显示,所提算法对于简单背景下的手势识别准确率可达99.79%,而对于复杂背景下的识别准确率为80.01%.

A new Bag-of-Features, namely hierarchical Bag-of-Features （H-BoF） was developed for the problem of computer vision based hand posture recognition. H-BoF captures image features＇ spatial information by dividing whole hand area into several sub-regions and projecting feature points onto horizontal and vertical directions respectively. To recognize hand postures accurately and rapidly, a simple yet effec- tive measurement based on histogram intersection kernel （HIK） was introduced which took full advantages of the H-BoF to classify hand postures. To improve the accuracy and efficiency of hand posture recognition against complicated background, a spectral based method with HIK--spectral-HIK was proposed to filter background features. Experimental results showed that this method achieved 99.79 % accuracy for uniform background samples, and 80.01% accuracy for complicated background samples.