In this section, we will discuss four popular low-level image segmentation methods and then compare the results obtained by those methods with an input image. The definition of good segmentation often depends on the application, and thus it is difficult to obtain a good segmentation. These methods are generally used for obtaining an over-segmentation, also known as superpixels. These superpixels then serve as a basis for more sophisticated algorithms such as merging with a region adjacency graph or conditional random fields.
Felzenszwalb's algorithm takes a graph-based approach to segmentation. It first constructs an undirected graph with the image pixels as vertices (the set to be segmented) and the weight of an edge between the two vertices being some measure of the dissimilarity (for example, the difference in intensity). In the graph-based approach, the problem of partitioning...
