Thanks!!
I made my own implementation, since complex image recognition would require to involve machine learning or similar prediction models.
What I do is to detect the contour (I'm using a LineRenderer to draw, so it is automatic) and downscale both images to process faster and have a higher fidelity % score.
With the contour I try to detect, rotate and stretch the image to look like Javi's image (I have some tracking beacons for it wink wink). From here I try to get 3 possible images, then I divide the drawing into 8 sections and try to see if there are pixels in the key areas.
Then I add more score if the pixels are equal for the rest of the drawing. I take the highest value of the three images.
If I can't detect the outline or the drawing is too strange, I simply compare the pixels one by one.