Take the 2-minute tour ×
Signal Processing Stack Exchange is a question and answer site for practitioners of the art and science of signal, image and video processing. It's 100% free, no registration required.

In my project, I have to digitize an ECG image taken with a normal camera (jpeg). For example, I have the following camera captured image:

Before

and I want to get something like this:-

After

and then the digitized data (x,y points), like in this video on digitization of ECG

I have no idea how to do it, so I searched and consulted several research papers. The general approach of the algorithms is like:-

  1. change to gray level image
  2. delete gridlines
  3. add missing points
  4. convert 2D image to 1D image

I am stuck with the second point, i.e., deleting the gridlines. I looked up some more references to do this and found histogram analysis might be helpful.

Can you please guide me on how to do this (I'm using MATLAB 2010)? Any help would be appreciated.

share|improve this question

migrated from stackoverflow.com Jan 3 '12 at 17:00

This question came from our site for professional and enthusiast programmers.

    
I asked basically the same thing here: stackoverflow.com/q/1657941/125507 –  endolith Jan 3 '12 at 21:07
    
@ Deepak,seems like you have implemented the matlab part of the code mentioned in the post ,can you kindly share the matlab part where you are converting the scanned image to 1D signal for interpolation –  user3278 Dec 10 '12 at 22:34
add comment

2 Answers

up vote 15 down vote accepted

Sorry, I use Mathematica, but it should be really easy to implement the idea into Matlab. I give the code anyway, so when my description is not detailed enough, you can get the rest from the code.

Basic idea is: you look at your image column-wise. Process every column of pixels separately. Note, that in the plot, I inverted the gray-values. So black is 1 and white is 0.
If you plot the (inverted) brightness-pixel-values you have basically only two situations. The first one is, when your column is not on a vertical grid line. The the plot looks like

The second situation is, where you are directly on a vertical grid line. Then the grid line influences the brightness of the whole column

But what you see is, that your dark EEG seems to be always the maximum. Therefore, the very complex algorithm is: Go through every column and take the position of the most black pixel.

img = ColorConvert[
   ImagePad[
    Import["http://i.stack.imgur.com/500Kg.jpg"], {{0, -20}, {0, 0}}],
    "Grayscale"];

Image[
 Transpose[Function[With[{m = Min[#]},
     Map[Function[{v}, If[v == m, 1, 0]], #]]] /@ 
   Transpose[ImageData[img, "Real"]]
  ]
 ]

Note, that I croped a bit of the right side of you image, since it was completely white there. The result is

Now you can join the points or interpolate them in every way you like and you get your EEG

share|improve this answer
    
@Patrick: I like the idea! –  Jonas Jan 3 '12 at 19:23
    
Note that this will only work if the gridlines are straight and parallel to the image axes –  endolith Jan 3 '12 at 21:09
    
@endolith, you have tried it, right?? Because here, it works when the image is not exact aligned. Btw, when the gridlines are not horizontal/vertical then the whole procedure of removing them is completely useless, since for a rotated image you would get wrong {x,y} values for the EEG. –  halirutan Jan 4 '12 at 12:11
    
@Patrick: I mean you would not get the correct x,y values if the data is not aligned with the coordinates of the pixels. –  endolith Jan 4 '12 at 17:41
    
@Patrick your algo worked :) and i have successfully extracted the signal but still having some difficulty in interpolating the signal (as i am very new to image processing) , kindly help me in how to interpolate the signal? thanks again :) –  Deepak Jan 5 '12 at 12:51
show 4 more comments

You have a colour image where the gridlines are red and the trace is black. so simply ignore any pixels which are red!

If you can't be sure the image is exactly aligned you could use the grid lines to calculate a skew (simply the slope of the image in pixels/pixel as you go right).

Then turning the black trace into a 1d value is simple. Start in the first column at the left edge and find the black pixel (or centroid of a small connected group of pixels) - the vertical position is your value.
Do that for each column in the image.
Where you have a missing value you will need to interpolate between the known values before and after.

For extra points you could set a limit on how much the trace can change from column-column to allow you to spot random dots or noise spikes.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.