摘要： 用深度图(点云)直接对曲面物体进行识别，将会加大曲面物体的数学表示的难度。该文提出一种对深度图进行自动分割的方法，基于微分不变量进行初始分割，得到初始的核区域，用区域增长法进行曲面片增长，将深度图分割成多个区域。该算法原理简单、易于理解和编程。通过一个米老鼠头部深度图实例证明了该方法的有效性。

关键词: 深度图像, 微分不变量, 区域增长, 自动分割, 物体识别

Abstract: An actual surface object consisteds of many surface patches. If surface object is directly recognized with range image, it is more difficult to represent the surface object. This paper proposes a method for automatic region segmentation algorithm of range image. The first original segmentation is made by differential invariant, and the range image is divided into many regions with region increase method. The algorithm is simple, and it is easy to understand and program. An example with range image of Mickey Mouse’s head proves the validity of the method.

Key words: range image, differential invariant, region increase, automatic segmentation, object recognition

中图分类号: 

• TP391