Recognize basic shapes in binarized image

edit

Generic shape recognition is a harder task and probably needs some training. If your shape set is limited (like in your images), it is easier to write a special detection algorithm for every shape.

For example: - distinguish shapes with a hole (empty circle, triangle) from filled shapes (filled circle, cross) using findContours()

• use houghCircles() on your contours to find out, if it's a circle. If it's not, it's a cross or a triangle

You can also play around with HuMoments()

You can also look the hierarchy of the contours.

If one contour is inside of another:

1. It could be an empty circle
2. It could be an empty triangle

If it isn't inside:

1. It could be a circle
2. It could a cross

Then you can also use approxPolyDP (approximate a curve or a polygon with another curve/polygon with less vertices) to distinguish between the two options.