用目标轮廓上点的相对几何位置关系对其形状进行描述,基于统计的思想,运用小波变换的方法,提出了一种新的轮廓描述子。与SuYang统计矩阵描述子相比,减少了计算量,大大提高了目标识别速度,并将这种方法应用到实时目标识别系统中去。实验表明,该方法可以快速提取目标轮廓对旋转、伸缩和平移具有不变性的特征,在实时目标识别系统中具有很好的表现。
Based on such a fact that the shape of a contour is described by point's relative geometry positions on it, this paper proposed a new contour descriptor through introducing the statistic and discrete wavelet transform (DWT) methods. Compared with Su Yang's descriptor, the new algorithm can greatly speed up targets recognition. Experiments demonstrated that the new descriptor is robust to contour noise and take the invariability of translation, rotation and scale variation.