How do I subtract the following two images and determine the number of pixels in a certain section of an image?

How would I subtract the following two images of the same size? My goal is to subtract and remove the purple section so I can determine the number of pixels in the gray section on the lower left (below the black line). My problem is that the purple section in both images aren't exactly the same, since they are different frames of a video. Would it be easier to just focus on the second image using some image segmenting or edge detection technique? Please help me out.

Answer 1

[Image Analyst]

Definitely not edge detection. What I'd do is use the Color Thresholder App on the Apps tab of the tool ribbon and segment in HSV color space. Then export the function to create a mask. Then you can use bwareafilt() to extract the largest blob or blobs of certain sizes.

[EDIT]

OK, since a lot of time has gone by and you still haven't responded, I'm assuming you couldn't quite get it. So I'm attaching a script that gives you the area.

area = 35389 pixels

Adapt as needed.

[Jialu Wang]

It works, thank you so much!

[Jialu Wang]

Hi again!

So I spoke to my prof about this and I'd still like to take an image difference because I'm trying to compare several different frames to an initial/background frame. My goal is to track the progression of the gray space and control its speed. I'd also like the code to word for several different substrates with different colours/saturation, so would it be possible to write a code that determines and adjusts the different threshold values? (right now it seems to be a constant) Please let me know, thanks so much.

[Image Analyst]

Well you could do a subtraction if you started out as all one color (pink) and then the gray moved in. You might want to do a median filter first to get rid of grainy noise. If the gray is always gray then you can probably threshold on the saturation channel regardless of what the color of the pink is. If both the pink and gray colors vary then you'll need to develop a segmentation algorithm for those. You can't use something like kmeans because you want it to work for any percent gray from 0% to 100% so there might not always be two classes in the image - it might be just one class, like if the gray has not moved in yet. But with kmeans you have to specify in advance how many classes there are. In that case you might use KNN (K nearest neighbors) or SVM and train it with an image halfway through, then use that segmentation on all the images.

[Jialu Wang]

Would it be better to segmnt both images so it's binary (black & white) and then do an image subtraction? Also I was wondering for the initial code you sent, how did you determine the threshold values for the channels based on histogram settings?

[Image Analyst]

No, don't subtract them after they've been binarized. If you have a good baseline/reference color or gray scale image, subtract that from each frame before doing color segmentation to form the binary image mask of the "test" frame.

[Jialu Wang]

Hi so I've tried subtracting the pink background image with a frame that has the gray moving in, but keep getting something that looks like this. What am I doing wrong? I'm not exactly sure how to apply a median filter--is that the issue? Should I convert to HSV space first?

Also what would a KNN or SVM code look like? Will the colour thresholding code you posted originally still work after I do the image subtraction?

Answer 2

[sangeetha N]

What to do if it was a video file? How to get the subtracted image?

If the objects are stationary,then after subtracting the input image and adaptive background image, I am getting number of objects as zero.How to solve this problem?