Abstract: A novel contour-based shape descriptor and shape matching method is proposed. Contour is extracted and points are evenly sampled on the contour. Distances between reference point and sample points, directions of contour on sampling points and the spatial relations between sampling points are used to represent the shape characteristics. Then scale, rotation and translation invariance on shape matching are obtained through maximum voting on different scales, orientations and positions. A similarity scoring scheme that integrates local and global features is proposed to perform object matching and detection. The experimental results show that ray shape model can be used in shape matching for objects with clear contours and detecting objects in complex images.

Key words: ray shape model, shape matching, object detection, maximum voting, transformation invariance

摘要： 提出一种新的基于轮廓的形状描述和匹配方法。提取物体的轮廓并在轮廓上进行等间隔采样，利用参考点到采样点的距离、采样点处的轮廓方向及采样点间的空间关系来直观地表达目标的形状特征；通过在不同尺度、方向和位置进行最大表决来获得形状匹配的尺度、旋转和平移不变性；提出了结合局部和整体特征的相似度评分机制来实现目标的匹配和检测。实验表明，形状的射线描述模型不仅能对具有清晰轮廓的目标进行有效的检索和匹配，也可在复杂的图像背景中检测目标。

关键词: 射线描述模型, 形状匹配, 目标检测, 最大表决, 变换不变性