一种多尺度自卷积快速算法  

A Fast Computational Algorithm of Multi-Scale Autoconvolution

作　　者：黄波[1,2] 赵晓晖[1] 庞怡杰[2] 时公涛[2] 陈东[2] 赵继印[1] 

机构地区：[1]吉林大学通信工程学院,吉林长春130022 [2]空军装备研究院,北京100085

出　　处：《电子学报》2013年第12期2430-2435,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61040043,No.61101213)

摘　　要：本文提出了一种MSA变换的快速算法.根据快速傅里叶变换理论,在MSA变换尺度的最小取值范围内,推导出最小基准变换尺寸,以取代同一尺度变换的不同变换尺寸,减少MSA变换计算次数;此外,在MSA变换尺度的最小取值范围外,利用MSA变换的对称性进行尺度范围映射,减小MSA变换尺寸,降低计算复杂度.利用典型数据,从时间效率和特征值精度对算法进行仿真分析验证.实验表明,所提快速计算方法在保证特征值精度一致的前提下,计算速度提高到3倍以上.A fast computational method of the multi-scale autoconvolution（MSA）transform is proposed in this paper .In or-der to reduce the times of MSA transform ,the method deduces the smallest benchmark transform size according to the fast Fourier transform theory ,and replaces the different transform sizes of the same scale transform within the minimum range of MSA transform scale .Then ,for reducing the computational complexity ,this method reduces the MSA transform by using the MSA transform sym-metry outside the range of MSA transform scale .Several experiments on the aspects of time efficiency and accuracy of eigenvalue using typical sample data are given .The results demonstrate that computation speed of the fast proposed computational method is three times faster than that of the original method while maintaining eigenvalue accuracy .

关 键 词：多尺度自卷积 (MSA ) 目标识别 仿射不变特征 仿射不变量 MULTI-SCALE autoconvolution(MSA) 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]