It is a good choice. To setup your SOM you need to define the input, the network 2D size, and then perform training.
you need to convert your image into vector(s), using net command you can define the size of your network and then use train() to train your network and obtain output. The output is the segmented image.