Before anything, A 'mask' is applied to the image to cover the picture in the middle

The image is then converted from RGB to HSV (Hue, Saturation, Value)

These can be seen below

We define color boundaries for the color we want to detect in HSV notation

These Boundaries are then used to create a mask

This mask is dilated to bring out the color.

This is done by performing convolution (with the current image) with a 5x5 matrix of 1s

The result is then bitwise ANDED with the HSV image to extract only the desired colors

Count the mask with the most number of white pixels

That color boundary used for that mask corresponds to the card type

A sample output can be seen below