The image has all the depth maps that are computed,
The image good filters these so you only get images with a good quality ( from the /quality topic)
The kf (keyframe) outputs a new image only when the viewpoint has moved significantly since the last keyframe,
and kf_good combines the two. (I'm not sure if it's keyframes from the image_good topic or the kf topic filtered to get only good keyframes).
So the good, kf and kf good images are exactly the same as the original images, you just get fewer of them.