有限等距常数(RestrictedIsometry Constant, RIC)是与有限等距性质(Restricted IsometryProperty, RIP)紧密结合在一起的一个参数。

一、RIC定义:

在前面的一篇RIP文章中其实已经提到了,这里在贴出该定义:

浅谈压缩感知(十七):测量矩阵之有限等距常数RIC的计算-LMLPHP

CS满足条件:S阶RIP性质只要要求0<δS<1就可以了,而RIC是指满足RIP的最小δS。

二、RIC计算:

RIC与特征值的关系:

浅谈压缩感知(十七):测量矩阵之有限等距常数RIC的计算-LMLPHP

浅谈压缩感知(十七):测量矩阵之有限等距常数RIC的计算-LMLPHP

三、MATLAB代码

function y=RIPText(x,t)
% calculete the kth Restricted Isometry Constant of measurement matrix x.
% Modified on March 13th, .
% Reference: Wei Dai, Olgica Milenkovic. Subspace Pursuit for Compressive Sensing
% Signal Reconstruction. IEEE Trans. ona Information Theory. vol., no.,
% -. May,
%calculate the Restricted Isomentry Constant of matrix constructed by
%random n columns of the original matrix x.
%created by Li Bo in March 16th,
%Harbin Institute of Technolgy
n=size(x,);%the numbers of column of original matrix x
Num=:n;%create a row of numbers from one to n with iterval
Com=combntns(Num,t);% all the combination of the selected t columns from total n columns
ma=size(Com,);% the number of combinations
delta=zeros(,ma);% a vector used to store the restricted isometry constant candidate
for i=:ma
b=x(:,Com(i,:));% new matrix constructed by the t selected columns
e=eig(b'*b-eye(t));%calculate all the eigen values of matrix b'*b-eye(t)
delta(i)=(max(abs(e)));%select the largest absolute eigen value as restricted isomentry candidate
end
y=max(delta);% select the largest one as restricted isometry constant

用此代码来验证矩阵其实是不现实的,原因解释如下,在程序运行过程中容易导致内存不足等问题。

浅谈压缩感知(十七):测量矩阵之有限等距常数RIC的计算-LMLPHP

四、参考文章

http://blog.csdn.net/jbb0523/article/details/44622819

04-25 10:46