Could you post some images here?

(I tried to pm you, but ResearchGate doesn't cooperate at the moment)

Here are a couple of the images.

This is the initial arrangement

The simplest solution can be as follows:

1. Segment the color image using public codes

2. Do morphological filtering using OpenCV or Matlab to remove noises and tiny holes

3. Count the number of regions, the dominant color of which is green or similar to a pre-defined color model, e.g. based on color histogram or Gaussian Mixture Model

The following link is a paper of flower segmentation. Its objective is different from yours, but it is an example of color image segmentation and histogram-based color model. Please ignore the complicated shape model they used because you do not need it.

http://www.robots.ox.ac.uk/~vgg/publications/papers/nilsback07.pdf

- Separate in RGB channels

- Convert from byte to short pixels

- Subtract R from G (1st image)

- Threshold at 20 (2nd image)

- (maybe a LocalMax, as Xiaomeng Wu suggested to get rid of straw that divides a marble)

- Calculate the area of each cluster, and decide what area constitutes a valid non-burying

(2nd image)

Thanks Lambert.. It looks like exactly what I wanted. I can calculate the area of the marbles w.r.t the initial pic to obtain % area buried. I'll try and obtain images like these. Unfortunately I have just started to use these image editing soft-wares, so could I trouble you in the future for a more detailed procedure?

Thank you Xiaomeng Wu, interesting article and definitely helpful in using the procedure in my case. As I mentioned, I am just beginning to use the imaging software and might trouble you again in the future for troubleshooting.

Thanking you all..

Sure, I will help if I can.

If you are only interested in one number (the % area buried), you could do a binary segmentation of the images before and after burying, as indicated above. And then just divide the average pixelvalue 'after' by the average pixelvalue 'before'.

(No need for LocalMax and no need to evaluate each separate cluster. But the distance of the camera to the scene should be kept constant.)

Mayank another approach so can see before and after and contrast between the two is by using positive and negative superimposition. This classically is the old fashioned way to make a bas relief photograph. The sequence is as follows:

1. Open your before photograph

2. Open you after photograph and then invert so that it becomes a negative

3. Select all of the after photograph

4. Go to the before and paste in the after as second layer, if using photoshop a new layer is created automatically

5. Decrease opacity of after layer to 50%.

6. If you then want to accentuate the differences increase contrast in usual way.

If both images were exactly the same you would see nothing until moved the individual layers in which case end up with bas relief.

By combining positive and negative you see before marbles as Green and after as Red so can then record movement and burial as Lambert and others have suggested.

Only thing is your camera must be in absolutely same position every time i.e. fixed otherwise get aberrations due to camera movement.

Looks interesting anyway!

Hi David Thanks for your input. Unfortunately I took these photos from my cell phone camera and therefore the distance, angle and luminosity would not be constant, even though I tried to keep it as such. However, I tried the method Lambert suggested and I got a good difference with a black background. Will use that to calculate the area in the initial and final photos. Thanks to all of you guys for the prompt and much needed help. :-)

P.s: I've attached a couple of pics of the threshold changed pics. Any further suggestions are most welcome.

Quote Mayank : "distance, angle and luminosity would not be constant".

In that case, put at least one object of known size in the scene just before you take a picture, so you can calibrate the areas (just a magnification factor). Preferably put three of those objects to account for different angles.

Yep Lambert.. I thought of that when I started analyzing the photos.. Thankfully still in the standardization phase.. will remember to put a scale and known object. Although I've measured the marbles, which are 1-1.3 cm in diameter, wouldn't that help?

RG is not letting me attach the photos as yet.. lemme try again.

Of cause you can use the marbles to calibrate, but at least one marble should be sufficiently visible in the 'buried' picture to determine its diameter. A scale is not necessary if you are only interested in a percentage.

Even more simple. Using Image J, you can subtract image A from image B. The resulting image will show those objects present in A and not in B and viceversa. If both images are similar (e.g. two adjacent frames of a video) the resulting image will show what has changed from one to the other.

No need to select one of the channels, or to segment the original images. If, afterwards you want to quantify the changes, you can segment the resulting image and measure the differences.

@Fernando: Yes, that would be nice! If only the pictures were taken at the same height/angle/lighting conditions, and if only the mice did not move around the straw.

Should you analyse the whole process of marble burying or just the final outcome? I assumed it was the second option (maybe wrong). If so, it seems quite simple. You take the original image (A, before putting the animal inside the cage) then you get the final picture (after removing the animal) you subtract both images, threshold and binarise the resulting image (A minus B) thus getting the surviving marbles plus small objects resulting from movement of the bedding, Then you can remove this using erode and reconstruct (these are available in Image J)

Applying a blur filter to both original images at the beginning might simplify the process. If you send me a couple of sample images I can try to refine the protocol and let you know.

Good luck

Fernando

@Fernando: Mayank has already posted two sample images in this thread, and also has described how the images are taken...

I rather like David Bryson's answer, but I suggest you register all Before and After images against a common Reference Image before any analysis. The freely available ImageJ / FIJI software has a "Projective Mapping plugin" that will allow you to register images quickly using manual selection of 4 points.

I'd suggest you put four small dots or permanent marks in the creases of the box just above the straw level. Then take a Reference image looking (nearly) straight down and centered above the middle of the straw. The four dots will then be very nearly a rectangle in the Reference image. Take your Before and After images as you have been, then use the Projective Mapping plugin to transform them to the reference view, then conduct your Before/After comparisons on the transformed images. The distortion of size and location caused by camera position changes will be mostly removed, though not entirely because the dots are not exactly in the same plane as the straw. -- Good luck.

I'm doing the same job now, I'm copying my results on the excel to draw graphs, because I took my results on the day of the test on the spot, my parameters that I took are: the number of ball completely buried, buried 1/2 ball, buried balls 2/3 and number of displaced balls are the other parameters are not important!?