How to detect peaks on gray background with Matlab FastPeakFind? - matlab

Question

I am testing the validity of the FileExchange project FindPeaksFast with different linewidths and backgrounds.
Test 1 is successful and the tool detects all peaks from 1px to 10 px.
However, Test 2 fails when testing to find peaks on the frame of an object plot i.e. an object (plot) on gray background.
The tool works well on white background.
Code
close all; clear all; clc;
f = figure;
hax = axes(f);
% Comment this out for Test 2
%zeroFigureDecorations(hax);
af = figure('Name', 'Do Not Touch');
x = rand(1,100);
y = rand(1,100);
linewidth=1;
plot(hax, x,y, 'LineWidth', linewidth);
I = getframe(hax);
I = I.cdata;
% https://se.mathworks.com/matlabcentral/fileexchange/37388-fast-2d-peak-finder
p=FastPeakFind(I);
% Input: 344x435x3 uint8
hold(hax, 'on');
plot(hax, p(1:2:end),p(2:2:end),'r+')
hold(hax, 'off');
function zeroFigureDecorations(ax)
axis(ax, 'tight');
set(ax, 'yTickLabel', []);
set(ax, 'xTickLabel', []);
set(ax, 'Ticklength', [0 0]); % http://stackoverflow.com/a/15529630/54964
colormap(ax, 1-gray(1024));
box(ax, 'off');
axis(ax, 'off');
end
Outputs in the following, and Fig. 1 shows that the function can detect something on lines when the background is white but not on correct locations.
Linewidth Output
10 166x1 double
1 844x1 double
Table: full axis decoration in Test 1
Linewidth Output
10 []
1 []
Table: no axis decorations, after zeroFigureDecorations(hax) in Test 2
Fig. 1 line as input (See Bla's answer) and its output,
Fig. 2 Output is wrong in Section 2,
Fig. 3 One more example that you cannot apply to function to simple curves,
Fig. 4 Section 3 Output is wrong, since not known how to apply the function on spectrograms
2 Test with bla's example data
f0 = figure;
hax0 = axes(f0);
d=uint16(conv2(reshape(single( 2^14*(rand(1,128*128)>0.9995) ),[128 128]) ,fspecial('gaussian', 10,2),'same')+2^4*rand(128));
imagesc(d, 'Parent', hax0);
I = getframe(hax0);
I = I.cdata;
p=FastPeakFind(I);
hold(hax0, 'on');
plot(hax0, p(1:2:end),p(2:2:end),'r+')
hold(hax0, 'off');
Output is wrong in Fig. 2
3 Testing with spectrograms
f3 = figure;
hax3 = axes(f3);
N = 1024*10;
n = 0:N-1;
w0 = 2*pi/5;
x = sin(w0*n)+10*sin(2*w0*n);
s = spectrogram(x);
spectrogram(x,'yaxis')
p=FastPeakFind(s);
hold on;
plot(p(1:2:end),p(2:2:end),'r+')
Matlab: 2016b
OS: Debian 8.5

Answer 1

You are not using the function correctly.
your code is this (verbatim):
f = figure;
hax = axes(f);
af = figure('Name', 'Do Not Touch');
x = rand(1,100);
y = rand(1,100);
linewidth=1;
plot(hax, x,y, 'LineWidth', linewidth);
I = getframe(hax);
I = I.cdata;
The matrix I is not a matrix that contain peaks like the function is intended to have. This is how it looks like:
imagesc(I);
Even if all you had were single pixels, that is not what the function is supposed to have, as it is said that the peaks point spread function needs to be larger than some # of pixels, and that they are assumed to be sparse. The function has a demonstration on a sample image that works fine.
Also , it's completely unclear what you even mean by peaks here.
EDIT:
Here's an example of how to use the function. First let's select random positions where we "create" the peaks:
I=rand(200)>0.9995;
This makes a binary matrix with only the points larger than 0.9995 selected (or having value 1). At each step you can imagesc(I) to see how I looks.
In real life, a camera will have some intensity in these points so we write:
I=I*100;
This is important as the peak by dentition needs to be a maximum value in its neighborhood. In real life, peaks are mostly not single pixels, they have some "width" or spread (this is also what the function says it deals with):
I=conv2(I,fspecial('gaussian',10,2),'same');
here, this spread is done by a "point-spread function" of a guassian of some width.
Lets add some 30% noise (note that after the last step the maximum value of the peaks is no longer 100, because it is spread to other pixels as well):
I=I+0.3*max(I(:))*rand(size(I));
Let's find peaks
p=FastPeakFind(I);
See how it did:
subplot(1,2,1);imagesc(I);
subplot(1,2,2);imagesc(I); hold on
plot(p(1:2:end),p(2:2:end),'r+')
In the function code, the example is doing what I wrote here in a single line. Note that there is an edg parameter, as this will not work on peaks on the edges of the image. This cab be solved by padding the image with zeros I think...