Splash water detection at swimming pool using thermal image

[EDIT: 20110525 13:50 CDT - reformat - WDR]

Greeting. I am developing an algorithm for splash water detection in the swimming pool. Here is my proposed algorithm.

1) extract image1 http://i54.tinypic.com/291i2oj.jpg

2) extract the black pixel from the image, as the splash water is represented with black colour.

3) once extract the black pixel, convert it to white in black and white image, save it as 'be4' image. http://i52.tinypic.com/almgd4.jpg

4) repeat step 1-3 for the next image. Save it as 'now' image after converting it to black and white. image 2 --> http://i52.tinypic.com/50d9c.jpg 'now' image --> http://i56.tinypic.com/t6b7l2.jpg

5) Conduct image subtraction between 'be4' and 'now' image. The result is imageResult. http://i52.tinypic.com/jijmt5.jpg

Problem: I do not know to use which criteria to differentiate the splash water from the other noise in the image [ like water wave ]. Here I attach my code.

a ='C:\Users\Ace\Desktop\splashTest\';
A =dir( strcat (a, '*.bmp' ));

for icount = 1 : length( A )

      %load the background
      %icount = 1;

      buf = strcat(a, A(icount).name);
      im1 = imread (buf);

      %imshow(im1)

      [y, x, z] = size(im1);

      %obtain black item  
          for i = 1:x
              for j = 1:y
                  RG = uint16( im1(j, i, 1) )+ uint16( im1(j, i, 2) );
                  if RG == 0
                      temp = 1;
                  else
                      temp = 0;
                  end
                  bw2(j,i) = temp;
              end
          end
       %figure, imshow(bw2);

       %load previous image whose black has been extracted and coverted to
       %white.
       be4 = bw2;

       buf = strcat(a, A(icount+1).name);
       im1 = imread (buf);    

       [y, x, z] = size(im1);

       %obtain black item  
          for i = 1:x
              for j = 1:y
                  RG = uint16( im1(j, i, 1) )+ uint16( im1(j, i, 2) );
                  if RG == 0
                      temp = 1;
                  else
                      temp = 0;
                  end
                  bw2(j,i) = temp;
              end
          end
       figure, imshow(bw2);

       now = bw2;

       imResult = imabsdiff( be4, now );
       %figure; imshow( imResult );

Sorry for my long-winded statement. I am very clueless now. Google also gives me no promising result. I hope Image Processing experts can suggest me some ideas. Thanks in advance.

Answer 1

[Sean de Wolski]

 bw2 = ~any(im1(:,:,[1 2]),3);  %your first set of for-loops, it'll be the same for the second where your define temp.

Instead of imabsdiff, you could use xor for logical images it'll be the same result just faster.

I guess I don't really understand what you mean by wave noise. Do you want to get rid of big black area, black areas that are touching the object in the previous image? Please describe what makes something "noise" as you refer to it.

[Eric]

if you view the imageResult file, my desired splash water is located at the top left corner. Other images are unnecessary. Is there any way or algorithm to differentiate the splash water near the top left corner from the other objects?

[Sean de Wolski]

Is the splash always going to not touch the border of the image? A connected components size threshold (bwareaopen) after a border clear (imclearborder) before you even compare images would give you just bigger black areas contained completely within the image.

[Eric]

Sometimes when the swimmer swims nears to the image border, the splash water will touch the image border.

Erm Do you suggest that I should use imclearborder, followed by bwareaopen before I compare 2 images?

[Sean de Wolski]

For the above image that would work. However, if the splash is on the border it wouldn't. It's very difficult for me (and everyone else) to give you pointers for generalization when we don't really know the scope of the project and every permutation of the images.

If you want the splash, define it: what makes it different from everything else: is it usually located in one spot (beneath the diving board?), is it always some shape that could be measured, it always a certain size, or distance from the pool edge, how about its magnitude of change between images? These are just a few of probably thousands of criteria you can use. Pick a few to build a profile of the splash, then the actual image processing is doable.