Single pixel event image processing in python

Question

I have had a lonng term problem for proceccing a CCD image of a Xray source, It is attached hereCCD image Then after theresholding with an arbitrary value, the thing I need to do is to Subtract the multi-pixel events from the image. I also need to count the number of pixels which belongs to single pixels.

*- By multi pixel I mean the pixels which has nonzero value in sorounding pixels.

I have a code working by using PIL.Image.open() to read it in to a list and do the analysis pixel by pixel! but I'm looking for a standard image prcessing routin for a more reliable and better results. I will appriciate if you can give me how to do it.

Cheers

So the result will be the number of pixels that are not touching any other pixels - correct? Do you need the image that has the multi-pixel events removed? — Mark Setchell, Sep 07 '16 at 14:38
Yes, the main idea is how many multi pixels do we have and how many single pixesl exist in there. Then I need to make the histograms which is easy. Also I have images which has many frames in there, like https://drive.google.com/open?id=0BwBIjnm0ViZOVlhaaFRNNERiY1U (100mB) which I need to do the process for each frame. — Alireza Honarfar, Sep 08 '16 at 11:30

Mark Setchell · Answer 1 · 2016-09-07T14:59:00.783

0

You can do this quite easily with ImageMagick which is installed on most Linux distros and is available for OSX and Windows. There are C/C++, Python, Perl, PHP, ,Net and other bindings available. I'll just do it at the command line here.

So, here is your start image:

First, let's threshold at an arbitrary 80%:

convert ccd.png -threshold 80% result.png

Now let's find all the white pixels that have no other white pixels around them:

convert ccd.png -threshold 80% -morphology HMT peaks:1.9 result.png

The technique is "Hit-and-Miss Morphology" and it is described here in Anthony Thyssen's excellent ImageMagick usage pages. Search on that page for "Peak" to find the exact paragraph.

Now let's count them:

convert ccd.png -threshold 80% -morphology HMT peaks:1.9 -format "%[fx:int(mean*w*h)]" info:
975

Let's check the multi-pixel events that we removed:

convert ccd.png -threshold 80% \( +clone -morphology HMT peaks:1.9 \) -compose difference -composite result.png

edited Sep 07 '16 at 14:59

answered Sep 07 '16 at 14:53

Mark Setchell

191,897
31
273
432

This is awesome, Thanks alot my friend. Then I have an other question, What I can do with multiple frame images, like : https://drive.google.com/open?id=0BwBIjnm0ViZOVlhaaFRNNERiY1U (100mB) – Alireza Honarfar Sep 08 '16 at 11:26
So, am I right that your TIF file just has 200 frames in it just like the one you gave me originally. If so, you can process any frame you like exactly as above but just give the frame number as well, i.e. `convert ccd.tif[0] -threshold...` for the first frame, and `convert ccd.tif[199] -threshold ...` for the last frame. – Mark Setchell Sep 08 '16 at 11:33
Your TIF file has some tags that ImageMagick doesn't recognise. You can suppress warnings like this `convert -define tiff:ignore-tags=4864,4865 ccd.tif[0] ....` – Mark Setchell Sep 08 '16 at 11:37
easy! thanks a lot again. I'm satisfied with your solution. – Alireza Honarfar Sep 08 '16 at 11:48
My pleasure. Good luck with your project! And remember, questions are free, so you can always come back and ask more ;-) – Mark Setchell Sep 08 '16 at 11:49
I came back again!, I wounder if it does the same job as http://docs.scipy.org/doc/scipy-0.15.1/reference/generated/scipy.ndimage.morphology.binary_hit_or_miss.html , So I will try to use this because I have some problem to get the ImageMagick into work on my system. – Alireza Honarfar Sep 08 '16 at 13:17
What system are you on? – Mark Setchell Sep 08 '16 at 13:48
Windows8.1 , Python(x,y) version 2.7 – Alireza Honarfar Sep 08 '16 at 14:00
I guess you would use something like this for the `scipy` version `structure1 = np.array([[0, 0, 0], [0, 1, 0], [0, 0, 0]]); structure2 = np.array([[1, 1, 1], [1, 0, 1], [1, 1, 1]])` though I don't speak Python. – Mark Setchell Sep 08 '16 at 14:46

Mark Setchell · Answer 2 · 2016-09-08T08:40:51.790

I had another attempt at this with the CImg library which is a header-only C++ image processing library that works on all platforms. That is rather nice since there is just one simple header file to copy into your project from the CImg website and you are finished!

See the CImg website here.

The code goes like this:

#define cimg_use_png
#define cimg_display 0
#include <iostream>
#include "CImg.h"

using namespace cimg_library;
using namespace std;

int main() {

   // Load input image
   CImg<unsigned char> im("ccd.png");
   int w=im.width();
   int h=im.height();

   // Create output image and fill with black (0)
   CImg<unsigned char> result(w,h,1,1,0);

   // Total number of white pixels with 8 black neighbours
   int t=0;

   // Threshold image at 80%
   im.threshold(255*0.8);

   // Find, and count, all white pixels with 8 black neighbours
   // Apply Dirichlet boundary conditions at edges - black virtual pixels at edges
   for(int y=0;y<h;y++){
      for(int x=0;x<w;x++){
         if((im.atXY(x,y,0,0)    !=1) ||
            (im.atXY(x-1,y-1,0,0)==1) || (im.atXY(x,y-1,0,0)==1) || (im.atXY(x+1,y-1,0,0)==1) ||
            (im.atXY(x-1,y,0,0)  ==1) || (im.atXY(x+1,y,0,0)==1) ||
            (im.atXY(x-1,y+1,0,0)==1) || (im.atXY(x,y+1,0,0)==1) || (im.atXY(x+1,y+1,0,0)==1)){
         } else {
            t++;
            // Paint output image white
            result(x,y)=255;
            cout << "DEBUG: " << x << "," << y << endl;
         }
      }
   }
   cout << t << endl;
   result.save_png("result.png");
}

And you can compile it with this:

g++ peaks.cpp -o peaks -lpng

Sample Output

This shows the x,y coordinates of pixels that exceed your threshold and have no immediate white neighbours.

DEBUG: 172,1
DEBUG: 287,1
DEBUG: 390,1
DEBUG: 396,1
DEBUG: 536,1
DEBUG: 745,1
DEBUG: 956,1
DEBUG: 72,2
DEBUG: 253,2
DEBUG: 516,2
DEBUG: 671,2
DEBUG: 680,2
DEBUG: 740,2
DEBUG: 811,2
DEBUG: 844,2
DEBUG: 228,3
DEBUG: 282,3
DEBUG: 351,3
DEBUG: 505,3
DEBUG: 551,3
DEBUG: 623,3
DEBUG: 638,3
DEBUG: 689,3
...
...
DEBUG: 797,252
DEBUG: 918,252
DEBUG: 125,253
DEBUG: 357,253
DEBUG: 870,253
DEBUG: 252,254
DEBUG: 941,254
1005

This looks good as well but I'm happier to stick to your first solution. Evenstill I didn't manage to get it in my python code. — Alireza Honarfar, Sep 08 '16 at 11:50

score 0 · Accepted Answer · answered Sep 20 '16 at 07:28

First of all I shall thanks @Mark Stechell for his hints and answers. Then I found my nswer by his hint about Hit-or-miss transformation. I found good information in wikipedia as : https://en.wikipedia.org/wiki/Hit-or-miss_transform

then in Python there is a ready made function for this methode which works for me and also for other paterning applications. you will find everithing in below link, it is straigh forward and easy. http://docs.scipy.org/doc/scipy-0.15.1/reference/generated/scipy.ndimage.morphology.binary_hit_or_miss.html

Cheers

Single pixel event image processing in python

3 Answers3