One of the most commonly used approach to segment a 2D image is the split and merge approach. In this paper, we are defining these two operations in 3D within the topological maps framework. This mathematic model of regions segmented image representation allows us to define these algorithms in a local and generic way. Moreover, we are defining a new operation, the corefining, which allows to treat big images. They are cut into small units, treated separately, then the result of each of them are combined to reconstruct the final representation. These three operations let us view efficient 3D segmentation algorithms, which is a difficult problem due to the size of data to treat.