The problem is your segmentation algorithm is no good. It's better to fix that than to try to repair a bad segmentation with bwconvhull() or activecontour() or something.
I don't know what you want, but if you want just the apple, you can simply threshold. Just get any color channel and see if it is not 255 (pure white).
I don't know what you did for kmeans, but if you want the whole apple you should use k=2, though like I said that is not necessary. But for what it's worth, I'm attaching my color segmentation by kmeans demo to this post.