Self-Organizing Map-Based Color Image Segmentation with k-Means Clustering and Saliency Map

Table 1

Parameters setting of different methods.

method	SOM-K(S)	k-means	JSEG
parameters setting	Preprocessing: normalize the variance of the variable to unity and its mean to zero.; Topology: 9×9 nodes, hexagonal topology.; Initialization and training: linear initialization, batch training.; Training epoch: rough training 4, fine tuning 2,; k-means: cluster number range: 2∼9, squared Euclidean distance, online update phase and batch update phase, initialize with selecting k observations from dataset at random.	Preprocessing: normalize the variance of the variable to unity and its mean to zero. Cluster number range: 2∼9, squared Euclidean distance, online update phase and batch update phase, initialize with selecting k observations from dataset at random.	default setting except 𝑞=600, 𝑙=1.