旋转图像的三维相关不变性  

Three dimensional correlation invariant based on spin images

作　　者：李世飞[1] 王平[2] 沈振康[2] 

机构地区：[1]空军预警学院训练部,湖北武汉430019 [2]国防科学技术大学ATR国家重点实验室,湖南长沙410073

出　　处：《红外与激光工程》2011年第12期2528-2534,共7页Infrared and Laser Engineering

基　　金：国防预研基金(9140A010107KG01)

摘　　要：三维不变特性的构建是三维目标识别过程中的一个核心步骤,旋转图像作为对三维目标曲面信息的一种很好的表达,满足了旋转和平移变换不变性,却不满足尺度变换不变性。为了构建一种在旋转、平移和尺度变换下的不变特性,首先定义了相关不变性的概念,然后使用Tsallis熵值对旋转图像进行了描述。在相关不变性的框架下,利用Tsallis熵值对旋转图像进行描述,在利用顶点坐标获取旋转图像之前,将所有的顶点进行归一化处理,得到了同时满足旋转、平移和尺度变换的相关不变性。理论证明和仿真实验验证了所构建的三维相关不变性。The structure of 3D invariant features is an intermediate but crucial step in 3D object recognition. As a descriptor of three dimensional object surfaces, spin-image has attracted a considerable amount of researchers＇ attention. Spin-images are rotation and transformation invariant, but they are not scale invariant. A terminology of correlation invariant was defined and a concise representation of spin- image was proposed by using the Tsallis entropy. So a three-dimensional object could be described using a vector which is composed of the Tsallis entropies. Before obtaining spin-images, all the vertexes were normalized so this representation of the Tsallis entropies, rotation, translation and scale translation were all correlation invafiant. Theoretical proof and numerical results show the proposed method＇s performance.

关 键 词：相关不变性 TSALLIS熵 旋转图像 尺度不变性 三维目标识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]