Script for indicating whether there is a crack in my image and how big the crack is problems.

I am creating a script that will detect a crack within an image tell me if there is a crack and if so how big it is and where it can be located but am struggling with a few parts of it, first of all I make it clear where the crack is by changing the image to red green blue layers and overlapping them showing a black and white image this does the job but my script doesnt say If there is a crack or not so I used an if else function to do this but seems to not be working either.

I then look at the different stats of the image using regionprops but here is where one of the first problem arises, how do I asses the stats its giving me the script I have gotten so far has gotten me nothing.

The last problem I have is I must find a way to convert pixel values to actual size I am using tihs code so far but where the distanceInPixels is I was thinking I could use one of the region props stats but which and how do i implemnt it into this?

%% Size converter from pixels to centimeters
 calibrationFactor =30/509 ;
distantceInCm =distanceInPixels * calibrationFactor;
areaInSquareCm = areaInPixels * calibrationFactor;

%% Import image
a= imread('hairline crack green.PNG');

%% seperating layers of the image
red=a(:,:,1);
green=a(:,:,2);
blue=a(:,:,3);


levelr= 0.2;
levelg=0.2;
levelb=0.3;

I1 =im2bw(red, levelr)
I2 =im2bw(green,levelg);
I3 =im2bw(blue,levelb);

%  imshow(I1)

  total=(I1&I2&I3);
      imshow(total);
      
 %% 
      ImageStats= regionprops(total, 'MajorAxisLength');

imgLengths = [imageStats.MajorAxisLength];
idx = imgLength > 15;
imageStatsFinal = imagestats(idx);
disp(imageStatsFinal)
%% Saying if there is a hairline crack
if isempty (imageStatsFinal)
    disp('The object does not have a hairline crack')
else
   disp('The object contains a hairline crack')
end

[Image Analyst]

Attach the ORIGINAL image, BEFORE you do "changing the image to red green blue layers and overlapping them", because I'm not sure that doing that is of any help in locating the crack. Your crack image you posted here looks very unnatural, like your segmentation routine was crummy. Post the original grayscale or color image, not this overly processed one. Or, do you not have a real image and all you have is this synthetic computer graphics simulated one?

[Thomas van Nedervelde]

Hi, I have other images which are more natural and real examples of cracks but they generate a lot of noise (white dots and such when in black and wihte) that is why I generated this synthetic one with very contrasting and strong colours. but an original image looks like this

I do the red green blue layering to allow for easier region analysis is this not effective?

[Image Analyst]

It's not. What I'd do (speaking with 40 years of image processing experience) is to first, put something behind your sample to remove the background clutter, such as a white cyclorama. Then I'd use the Color Thresholder app to find parts of the image with high saturation and in the red hue range. Use that mask to get the area, and if the area is big enough, there is a red dye leak. You could also look at the HSV values or LAB values within the mask to determine the overall severity of the leak.

[Thomas van Nedervelde]

Thank you for that answer I will most definetly look into the color thresholder app and letting it indicate my area of higher red saturation levels, do you think my way of using a background object such as a ruler with known size would be suitable to find genuine area sizes? and using a calibration factor like this?

 calibrationFactor =30/509 ;
distantceInCm =distanceInPixels * calibrationFactor;
areaInSquareCm = areaInPixels * calibrationFactor;

The main problem I'm facing with this is being unable to see what it would look like automised, like being able to run a script and it assesing the image automatically, is this viable do you think?

below attached is an image with less background clutter do you think this would be proccesable or would the background have to be fully cleared/ whited out?

Thank you so much for your help it is my first time trying to devolop a script that analyses images and I am struggling with a lot of part of it.

[Image Analyst]

This looks totally different. Let's say you were to try to find dark red pixels. Well there are some panels in the back that would also be found. So I think the best is for you to just lasoo the area you want to analyze manually to make sure no other dark red things get included and determined to be cracks when they're not. See attached freehand masking demo.

[Thomas van Nedervelde]

Thank you I wasnt to sure on what masking was but now understand it, how would I make my script analyse the mask though? would I go back to assesing the RGB values or would you suggest using the colour treshholding app? and generating a script for that or can that not be automise?

In your many years within this industry have you seen many scripts like this that excute something similar? If so do you know about any other threads that might be helpfull?

Kind regards

[Thomas van Nedervelde]

Hi image analyst I have been playing around with the masking script you attached and I was wondering why when I use my own image it only creates a binary mask of the region and it doesnt create any of the other masks?

folder = fileparts(which('p.jpg')); % Determine where demo folder is.
baseFileName = 'p.jpg';

surely these are the only two lines I have to change since the file is not called upon anywhere else?