Unexpected behaviour of function findpeaks in MATLAB's Signal Processing Toolbox - matlab

Question

Edit: Actually this is not unexpected behaviour, but I still need a solution.
findpeaks compares each element of data to its neighboring values.
I have data which contains peaks which I detect with the function findpeaks from the Signal Processing Toolbox. Sometimes the function seems not to detect the peaks properly, when I have the same value twice next to each other. This occurs very rarly in my data, but here is a sample to illustrate my problem:
>> values
values =
-0.0324
-0.0371
-0.0393
-0.0387
-0.0331
-0.0280
-0.0216
-0.0134
-0.0011
0.0098
0.0217
0.0352
0.0467
0.0548
0.0639
0.0740
0.0813
0.0858 <-- here should be another peak
0.0858 <--
0.0812
0.0719
0.0600
0.0473
0.0353
0.0239
0.0151
0.0083
0.0034
-0.0001
-0.0025
-0.0043
-0.0057
-0.0048
-0.0038
-0.0026
0.0007
0.0043
0.0062
0.0083
0.0106
0.0111
0.0116
0.0102
0.0089
0.0057
0.0025
-0.0025
-0.0056
Now the findpeaks function only finds one peak:
>> [pks loc] = findpeaks(values)
pks =
0.0116
loc =
42
If I plot the data, it becomes obvious that findpeaks misses one of the peaks at the location 18/19 because they both have the value 0.08579.
What is the best way to find those missing peaks?

Answer 1

If you have the image processing toolbox, you can use IMREGIONALMAX to find the peaks, after which you can use regionprops to find the center of the regions (if that's what you need), i.e.
bw = imregionalmax(signal);
peakLocations = find(bw); %# returns n peaks for an n-tuple of max-values
stats = regionprops(bw,'Centroid');
peakLocations = cat(1,stats.Centroid); %# returns the center of the n-tuple of max-values

Answer 2

This is an old topic, but maybe some are still looking for an easier solution to this (like I did today):
You could also just substract some very small fixed value from all values on a plateau, except from the first value. This causes each first value on a plateau to always be the highest on the respective plateaus, causing them to be included as peaks.
Just make something like this part of your code:
peaks = yourdata;
verysmallvalue = .001;
plateauvalue = peaks(1);
for i = 2:size(peaks,1)
if peaks(i) == plateauvalue
peaks(i) = peaks(i) - verysmallvalue;
else
plateauvalue = peaks(i);
end
end
[PKS,LOCS] = findpeaks(peaks);
plot(yourdata);
hold on;
plot(LOCS, yourdata(LOCS), 'Color', 'Red', 'Line', 'None', 'Marker', 'o');
Hope this helps!

Answer 3

Use the second derivative test instead?

Answer 4

I ended up writing my own simpler version of findpeaks, which seems to work for my purpose.
function [pks,locs] = my_findpeaks(X)
M = numel(X);
pks = [];
locs = [];
if (M < 4)
datamsgid = generatemsgid('emptyDataSet');
error(datamsgid,'Data set must contain at least 4 samples.');
else
for idx=1:M-3
if X(idx)< X(idx+1) && X(idx+1)>=X(idx+2) && X(idx+2)> X(idx+3)
pks = [pks X(idx)];
locs = [locs idx];
end
end
end
end
Edit: To clarify, the problem arose, when I had a peak which was exactly between two sample points and those two sample points had coincidentally the same value. It only happend a couple of times in more than 10.000 cases.

Answer 5

The behavior that you describe is a known bug in versions of MATLAB prior to R2010b. The minimum example is
findpeaks([0 1 1 0])
which returns [], while
findpeaks([0 1 0])
returns the (position of the) peak.
The bug has been fixed in R2010b and later, see the official Bug Report. With that fix, findpeaks returns the rising edge of "peaks with repeated values" (which I would call plateaus).