First up, it finds the region the two image rectangles intersect, then it iterates through each pixel in that region, and compares the alpha values of each image of each pixel. If neither has an alpha value of 0, they are both considered 'solid' and therefore colliding.