as mentioned in the paper, based on the predefined "percentage of CHs", some nodes select their-self as CH candidate. then, they send a message to inform the nodes around them to being cluster heads. when a typical node receive multiple messages it select the closest CH based on the signal strength. this way, usually all nodes receive at least one message. however, if a node don't receive any message, it become a CH automatically and send information directly to the sink node.

in leach protocol CH will be selected based on the maximum energy of node and other nodes which are in the range of CH will be the member of cluster.

Thanks everyone for answer. @Mohesn mafakheri  

If a node doesnot recceive any msg from a CH. Then it becomes a CH? But number of CH are k (optimal) in each round. Suppose we have taken k=5. Then we can not have more than 5 CH in a round. If 5 nodes are already CH s in a round. Then 6th node can not be CH in the round.

LEACH talk about k cluster heads. but it cant and wont guarantee k cluster heads. The reason is that "Leach uses probabilistic threshold value".  The original leach protocol elect cluster heads using random number (not energy of the nodes). Any node wishing to become cluster head in a particular round chooses a random number between 0 and 1, if the number is less than a threshold value, it is elected as cluster head. As a result, a node having very low energy may be elected as cluster head at the expense of high energy node. If you want to elect an optimal (it is near optimal, not exactly k cluster heads in each round) number of cluster head, then centralized approach need to be used in which base station elect the cluster heads. Also, those nodes which are not part of cluster are not cluster head. but they act as normal node similar to member nodes of a cluster. Cluster head collect data. but such nodes which are not left alone, only transmit their own data (not the data of other nodes).  

Thanks Mians

you are welcome :)