基于自适应模板半径的形状上下文描述子在手势识别上的应用

Hand Gesture Recognition Using a Modified Shape Context Descriptor Method Dased on the Adaptive Template Diameter

提出了一种基于自适应模板半径的形状上下文描述子描述图像轮廓特征的手势动作模式识别方法,并探讨了这种方法在基于视觉的汉语字母手势识别中的应用。首先,在计算手势轮廓各点的形状上下文时,提出了一种距离直方图法求取自适应的模板半径的方法,并以χ2统计量计算各轮廓点之间的匹配距离,再以改进Hausdorff距离计算各手势轮廓间的匹配距离,最后用最近邻法进行分类识别。对来自50个实验对象的手势图像分类识别的实验结果表明,该方法对30个汉语字母手势图像的正确识别率达到了74.3%,明显高于基于不变矩及傅里叶描述子特征提取方法的识别结果。

A modified shape context descriptor method based on the adaptive template diameter was proposed for pattern recognition of hand gestures. In this method, distance histogram is used to calculate the adaptive diameters of templates, which is adopted to compute the shape contexts of the points on hand image silhouette.x2 statistic value is computed to denote the distance of two different points, and modified Hausdorff distance to represent the similarity distance of two hand image silhouettes. Nearest neighbor classification method are then used for hand image recognition. With the proposed method, a verage correct classification recognition rate was 74.3% when pattern recognition experiments were implemented on 30 classes of alphabetic gestures of Chinese sign language from 50 subjects. The experimental results demonstrated that the proposed method outperformed the existing classification methods based on invariant moments or Fourier descriptor.