I'm not sure why you say that the output is not correct. It seems fine to me.
See, if you run k-means using the squared Euclidean distance (as you did) the clusters will be biased towards spherical shapes. Unfortunately for you, one of the clusters in the image is not spherical. You can see that each spherical cluster has a unique colour, but the cluster that isn't spherical doesn't.