Representations of an object in an image generated by an imaging apparatus can comprise two or more separate sub-objects, producing a compound object. Compound objects can negatively affect the quality of object visualization and threat identification performance. As provided herein, a compound object can be separated into sub-objects. Three-dimensional image data of a potential compound object is projected into a two-dimensional manifold projection, and segmentation is performed on the two-dimensional manifold projection of the compound object to identify sub-objects. Once sub-objects are identified, the two-dimensional, segmented manifold projection is projected into three-dimensional space. A three-dimensional segmentation may then be performed to identify additional sub-objects of the compound object that were not identified by the two-dimensional segmentation.