采用级联仿射不变函数的快速平面形状识别(英文)  

Fast planar shape recognition using cascaded affine invariant function

作　　者：史思琦[1] 石光明[1] 陈旭阳[1,2] 齐飞[1] 

机构地区：[1]西安电子科技大学智能感知与图像理解教育部重点实验室,陕西西安710071 [2]空间电子信息技术研究院,陕西西安710000

出　　处：《红外与激光工程》2012年第9期2534-2542,共9页Infrared and Laser Engineering

基　　金：国家自然科学基金(60736043,60805012,6103304)

摘　　要：现有的基于小波变换的形状识别算法具有很高的计算复杂度,难以满足许多实时应用的要求。文中提出了基于级联仿射不变函数的快速形状识别算法,用于识别仿射变换下的含噪目标。利用目标轮廓的小波变换可以得到一组仿射不变函数,并进一步构造出级联仿射不变函数。为了保证级联仿射不变函数的平移不变性,预先对轮廓的起始点进行了有效配准。从而通过级联仿射不变函数的内积,方便地度量出目标形状的相似度。与现有基于小波的识别算法相比,所提出的算法具有很低的计算复杂度,其所需CPU时间仅为其它算法的1/7。实验结果验证了该算法的有效性和起始点配准的准确性。Existing shape recognition methods based on wavelet transform are of high computational cost,so they are not suitable for some real-time applications.A cascaded affine invariant function（CAIF）-based fast recognition algorithm was proposed to recognize the noisy shape under affine transform.The CAIF was constructed by combining a set of affine invariant functions,which were derived by the wavelet coefficients of the shape contour.To ensure the shift invariance of CAIF,the starting point of shape contour was pre-aligned effectively.In this way,the shape similarity could be measured conveniently by the inner product of CAIFs.The proposed algorithm was of low computational complexity and spent only one seventh of the CPU time consumed by other wavelet-based recognition algorithms.Experimental results verify the alignment accuracy and the efficiency of the proposed algorithm.

关 键 词：形状识别 仿射变换 离散小波变换 级联仿射不变函数 起始点配准 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]