How to determine the most accurate value of thresh holding 'level' used to segment very fine cracks in Matlab using the im2bw(I2,level) function ?

I am currently using 'im2bw(I2,level)' function in Matlab(see attached code below)for 3 different images. At the moment inbuilt Matlab Otsu's method 'graythresh(I)' gives poor result while setting level in im2bw(I2,level) by trial and error is too hectic.

How can I obtain the most correct value for 'level' to segment most/if not all of the very thin/fine cracks for each image (see attached images)? Can anyone help with how I can go about doing this for each image? 

    ..............................................
          I=imread('L1.jpg');
          figure,imshow(I)
          I2=imcrop;%Crop image to contain only steel surface containing cracks
          figure,imshow(I2)
          title('Original grey Image')

        %%Binarize
        level = 0.7303; % thresh holding level by trial and error too hectic!!!
        %level = graythresh(I);% Matlab in-built global thresh holding based on Otsu's method returns NOT SO GOOD           
                               %VALUES FOR 'level' are obtained from this!!!  
        binaryImage = im2bw(I2,level);
        figure,imshow(binaryImage)
        title('Binarized image') 
    ..........................................................

Images:-

Kind regards. BB

[BB BSB]

Please, I need help from image specialist to improve the binary segmentation. Any contribution will be much appreciated!