Better way of "binary image multiplication"

Hello, Question title may be unclear, but I dont know "how to ask" this question in better way.

Long story short - I am doing Landsat image change detection. At the moment, I used two years - 1990 and 2013. In year 1990 I have 3 images, in year 2013 - one.

I did my thing and got 3 binary images, representing locations, where change has happened. The Idea - i combine these 3 binary images, and I would get one clean binary difference image - I would eliminate things like clouds in one or another image. What I did now -

totalDifferences = diff1 .* diff2 .* diff3;

But there might be a flaw. Imagine regions like these:

     diff1           diff2
  [0, 0, 0, 1;    [ 0, 0, 0, 1;
   0, 0, 0, 1;      0, 0, 0, 1;
   0, 1, 1, 1;      0, 1, 1, 0;
   0, 1, 1, 1];     0, 1, 0, 0];

This would produce totalDifferences the same as diff2, cause of multiplying with 0. And with 3 real images, it would shrink my valid mask even more, maybe because of shadows, or something like that.

My thought was - keep those regions (not plain pixels), which intersects, but eliminate total outliers. Any ideas, if this is do-able and maybe how?

Answer 1

[Image Analyst]

You need to first and the images using the & operator. Then you need to use imreconstruct(). See my attached demos.

[Maris Samats]

Yes, that is exactly what I need! Thank you.