To explain with an example:
In one class, I showed students a plot that I wanted them to draw. They saw it as too short and tried to draw the same. Some were very similar, some were less similar, and some were unrelated. I want to find the similarity ratio between them and grade them accordingly. I discovered that I can use algorithms such as ORB(oriented fast and rotated brief), SSIM(Structural similarity index measure), and VGG16(Visual Geometry Group) to find this similarity ratio most accurately (if you know a better method, I would like to hear it). First of all, I determined some layers to optimize the algorithm I will use. These are a squared area, an area with only defined axes, and an empty area. All images have the same size area. When I used the squared area, I expected to see very low similarities between the undrawn version and the drawn version . but the similarity rate was high. This time I tried drawing in an area with only axes . The similarity between the undrawn version and the drawn version was still high. Moreover, I drew an irrelevant semi-circle and the similarity was still very high. I tried just drawing on a white area but failed again. While I have had such unsuccessful results even with my drawings on the computer, I cannot expect a successful result from a paper drawn on paper and scanned. How can I configure it? Should I try a different method?