fft and lpc
in [Matlab]
|
From: Vassilis Papavassiliou on 8 Dec 2009 11:17 suppose I have a dark image A (hxw pixels) which contains bright vertical lines. It's projection y <y=mean(A,1)-mean(A(:)); seems to be a periodic signal. apply a hamming window (explained later) < y = y.*(hamming(size(y,2))'; Getting its spectrum <F =abs(fft(y)); I found the first lobe at position (frequency) f. Thying to estimate the smooth spectrum of this projection by using an lpc model of 3rd order (that's why the hamming is used) <[coeff g]= lpc(y,3); Suppose I have a pole with angle c1 and amplitude a1. <roo = roots(coeff); ang = angle(roo); ampl = abs(roo); <c1=ang(1); a1=ampl1; if I rezise the image by a factor 0.2 and repeat the same process, I get another spectrum Fnew which is analogous to the original one F. But the new angle a2 is not. Can anybody explain this? Note that I do not want to reconstruct the signal y or y1 (the new projection). I suppose that lpc (autocorreellation method) by just one pair of poles is not robust. |