It takes a cloud of points as input, and gives a set of shapes that approximate that cloud, noise filtered out. So component is one of those shapes. In case of

It takes a cloud of points as input, and gives a set of shapes that approximate that cloud, noise filtered out. So component is one of those shapes. In case of PCA, shapes are ellipses. For SOM, shapes could be more complex like zigzag. For k-means, shapes would be like Voronoi cells.