# Smoothing only peaks in a Curve

14 views (last 30 days)
Anirudh on 21 Feb 2014
Commented: Anirudh on 31 Mar 2014
Hello,
The curve I have is shown in the image. Is there a way I could smooth only the peak, which is around the X=800 mark, ideally keeping the rest of the data as the raw signals.
In case I decide to join a raw and a smoothed curve (using a filter), the problem then I face is the joining point of the raw curve and the smoothed curve, which shows a step (undesired).
I would like to ideally keep the raw data till X=720 (point where constant rise starts), then continue with the raw data through the constant slope, i.e from x=720 to the point where smoothing of the peak starts, perform a smoothing with a factor and rejoin back with the raw data once the smoothing is done (keeping that step in mind again).
I was talking to one of the correspondents and it was nice of him to help me out on this thread, http://www.mathworks.com/matlabcentral/answers/116391-smoothing-only-a-part-of-a-curve#answer_124995 but I'm kinda needing an urgent answer, and if possible today.
Could anyone out there help me out on this one?
Regards,
Anirudh
Walter Roberson on 9 Mar 2014

Nitin on 22 Feb 2014
Did you check the smooth function?

Image Analyst on 22 Feb 2014
Just smooth the whole thing, say with conv() to do a sliding average or with sgolay to so a sliding polynomial Savitzky-Golay filter. Then replace the original pixels with the smoothed one in the range you want
smoothedSignal = conv(originalSignal, ones(1,9), 'same'); % or whatever
newSignal = originalSignal; % Initialize
% Replace region of interest with smoothed values.
newSignal(index1:index2) = smoothedSignal(index1:index2); % You determine the indexes in some way.
Anirudh on 31 Mar 2014
Image Analyst, Thanks for writing back :).
I used the conv() function, smoothed the curve and replaced only a certain smoothed part (say X=750 to X=820), in the raw signals. I then smoothed this entire curve once again to eliminate the steps at X=750 and X=820.
But this is not a very convincing method as I'm deviating from the original values by more, so I'm still on the lookout for a better solution to this.
Thanks, Anirudh