How do I compare a sample pocket image with test images to test for sizing errors?

I am a newbie to matlab and image processing in general however my sister asked for me to biuld a project for her.

The project goes as follows

Given an image of a samples of the pocket design from a distance x, determine the size estimation of the sample pocket as well as compare them to the size estimations of all test images in the test set taken from the same distance x.

I was think If could detect the edges of the pocket and then somehow compute a pentagone from the picture and compute its size and compare it to all other pictures than I could test how much difference there is to determine the error.

I have also heard about sementic segmentation using nerual networks and am wondering if that would be a better option than this.

Regardless, I am student and am fully aware that this project is above my level but I still don't wish to let my sister down.

Any help would seriously be appreciated.

sample image:

image to test:

Answer 1

[Image Analyst]

Some things to try

Turn it into gray scale or use the red image

[r, g, b] = imsplit(rgbImage);

whichever one has more contrast.

then try imgradient, a canny filter, stdfilt, and adapthisteq. Then whichever one is best, threshold it. Might have to call bwconvhull and then bwareafilt. You might be able to get a big pocket-shaped blob from doing all that. You can certainly try SegNet but you'd need probably at least 50 images to train it on. With traditional image analysis algorithms you don't need any, or just the 3 or 4 you use during algorithm development.

If you need anymore help, attach the image it's not working on along with your code.

[Syed Mohammad Abuzar]

But how do I find the area of the blobs and compare them?

[Image Analyst]

That would be regionprops. See my Image Segmentation Tutorial for a full demo.

props = regionprops(binaryImage, 'Area');
allAreas = [props.Area]

See my Image Processing Tutorial in my File Exchange:

Image Analyst's File Exchange