Figure 1

Working principle of phew . The main steps of the algorithm are visualized on a 1D density field (first panel). The segmentation into peak patches is shown in the second panel. Based on the relevance of a peak (peak-to-saddle ratio) we decide whether a peak represents ‘noise’ or substructure. Irrelevant peaks are merged through their highest saddle points (third panel). The surviving objects are labeled as Level 0 clumps and denote the finest level of substructure. The substructure is merged based on a saddle threshold (third panel) into parent structure (fourth panel).