isaacgerg/matlabhyperspectraltoolbox

MATLAB Hyperspectral Toolbox

matlabhyperspectraltoolbox's Issues

hyperHfcVd

I was looking at the function hyperHfcVd in the hyperspectral toolbox. It was giving me zero dimensions in one case. There are a few things that look incorrect to me, but I’m not entirely confident in these:

1. lambdaCov = flipud(eig(cov(M’))); and similar for lambdaCorr: I’d replace flipud with sort. I think it doesn’t matter whether they’re increasing or decreasing, but lambdaCov and lambdaCorr need to have analogous order, and eig doesn't guarantee the order, I believe. At least I'm pretty sure I've seen cases in which the eigenvalues were not ordered.
2. I think the corr function returns a matrix of correlation coefficients, with maximum value 1, minimum -1. I think something like

    m = mean(M,2);
    C = cov(M’);
    lambdaCov = sort(eig(C), ‘descend’);
    lambdaCorr = sort(eig(C+m*m’));

is appropriate, but I’m not sure of the normalization to get unbiased values for lambdaCorr.

1. sigmaSquared = (2*lambdaCov(x)/N) + (2*lambdaCorr(x)/N) + (2/N)*lambdaCov(x)*lambdaCorr(x);
   I think this should be

sigmaSquared = (2*lambdaCov(x)^2/N) + (2*lambdaCorr(x)^2/N) - (4/N)*lambdaCov(x)*lambdaCorr(x);

Note the squares, the minus sign, and the 4. Look at Chang and Du, the text from Eq. 7 to Eq. 9. The variances in Eq.7 are 2*lambdaCov^2/N, and 2*lambdaCorr^2/N. In addition, Eq 7 has -2*Cov, and the covariance is 2*lambdaCov*lambdaCorr/N. In fact sigmaSquared could be written as
(2/N)*(lambdaCorr-lambdaCov)^2, if all of that is correct.

I am not terribly confident in those changes, since it still gives odd results for my cases.

Clay

HyperDemo_mams_RTI_data

thanks for your time. I wanted to ask you in hyperdemo_mams_RTI because in the first "for =1:p" you do this. "p" is the number of bands, you should find many abundance map as there are endmember "q"?
because every time you try a number of endmember getting bigger up to "p = 126"?
many thanks

not working hyperdemo

I wanted to ask, because when I start with hyperdem0.m at some point it stops matlab? in particular when running hyperresample. I use matlab R2015.