I hope this paper can help in your decision. Please take a look on table 1, here you can find the number of clusters used in different studies of the K-means algorithm. 

you can use fuzzy clustering approach. you can find the details from my publication.

Your choice of the algorithmic tool that you might use mainly depends on the problem that you have in hands.

For example if you have to deal with a problem that involves a lot of uncertainty maybe your best clustering tools that you might use are fuzzy clustering and/or learning vector quantization algorithms.

In my opinion k-means is the best.

in many research i found , they used K-means for clusturing images or K-Centroîd can be used also

In my opinion the best method is some incremental algorithm (see e.g. Bagirov et al., Pattern Recognition 44(2011), or Scitovski, Computers & Geosciences 59(2013), or Morales-Esteban et al., Computers & Geosciences 73(2014)).

An important advantage of all incremental algorithms for searching for an optimal partition lies in the fact that one obtains an optimal partition for each k ≤ kmax, where kmax is given in advance. This allows the estimation of the appropriate number of clusters in a partition by using various well-known indexes (see e.g. Gan et al., Data Clustering, SIAM, 2007, or Vendramin et al., On the comparison of relative clustering validity criteria. In Proceedings of the SIAM International Conference on Data Mining, pp. 733–744, 2009)

It is very difficult to say that a unique clustering algorithm can able to clustered any kind of image. It depends on the distribution of the pattern. If the pattern is circular or Gaussian distribution with limited spread then K-means is best. If the pattern is not circular or non-Gaussian then apply multi-seed based clustering technique. See the reference: "A novel multi-seed non-hierarchical data clustering technique", IEEE Trans. on Systems, Man and Cybernetics, Vol. 27, No. 5, pp. 871-877, 1997.   

There is an automatic clustering method I want to recommend you.

Automatic histogram-based fuzzy c-means cluster'ng method ==> AHFCM method

This method that I generated is an automatic clustering method witch does not need to be initialized with the number of clusters to work. the method determines the number of clusters optimally with the histogram-based method. and the user Interference is zero.

Good luck.

I used K-means and FCM in the following publication. You can observe their results and efficiency.

Article Automatic building extraction in dense urban areas through G...

In my opinion,  k-means algorithm is the best. Further, it is important the representation of information associated with each pixel and ultimately most important is the similarity or distance function used in thek-means  algorithm.

You can mix K-means then AHC