Post Processing

Each name says what goes, so read it as its opposite: Remove labels (smaller) keeps the larger events. In the reference's examples the four Remove labels operations return a coded image, and the two Remove objects operations are shown on 8-bit masks and return a mask. The surviving events keep their numbers. The Re-Label entry names Remove labels as a source of missing labels and recommends Re-Label (missing)Loading... afterward.

Hole filling and small-object removal on masks belong to the Morphological group: Fill holesLoading... (setting: Allow fill on border), Fill small holes and Remove small objects (from mask) (setting: Area threshold (px) for both).

Source: StrataQuest 8.1 Engines Reference, ch. 6.9 (pp. 284–303); ch. 6.8 (pp. 227–255); ch. 6.6 (pp. 195–198).

Cleanup comes at two levels, and it helps to keep them apart. Pixel-level cleanup changes a mask's shape with mathematical morphology. Object-level cleanup keeps or deletes whole objects, so every survivor stays exactly as detected. StrataQuest's Morphological group works at the first level, and the Post Processing group at the second.

Morphological shaping

In mathematical morphology generally, a small probe called a structuring elementLoading... determines what changes. Erosion keeps only the points where the probe fits inside the object, so the object shrinks. Dilation adds every point the probe covers as it slides along the object, so the object grows.

The reference gives a concrete case: a square of side 10 dilated by a disk of radius 2 becomes a square of side 14 with rounded corners, and eroded by the same disk it becomes a square of side 6. In StrataQuest, Erode, Dilate, Open and Close each take Kernel radius (px) and Kernel shape (default Disk).

A morphological openingLoading... (erode, then dilate) removes small objects and protrusions, and a closing (dilate, then erode) fills small holes and gaps, each with the same kernel. Both reshape every object they touch: an opening rounds the corners of the survivors too.

Distance Transforms & Watershed

Separating touching objects isn't a post-processing operation, but it sits close by in the pipeline. In image processing generally, a distance mapLoading... gives each foreground pixel its distance to the nearest background, and its peaks mark object centers. A watershedLoading... grows regions from those centers and draws a boundary wherever two meet. StrataQuest has its own distance map and watershed engines for this.

Connected Component Analysis

Object-level work needs objects first. Labeling gives each connected group of foreground pixels its own number, and StrataQuest's Label operation offers 4-connectivity (the default), which joins pixels that touch horizontally or vertically, or 8-connectivity, which adds diagonals. The choice determines whether two pixels touching at a corner are one object or two.

Once objects exist, filtering by area is the object-level counterpart of an opening. The reference calls Remove small objects (from mask) an area opening: it removes small objects, and unlike a morphological opening it leaves the shape of every survivor untouched. Remove labels (smaller) does the same job on a coded image.

Pipeline Ordering

The reference stresses that order matters inside the Basic Operations Module, because an operation can use only the outputs of operations above it. It recommends no fixed sequence. A sequence that follows from what each step does: clean the mask (fill, then remove specks), label it, delete unwanted events, then re-label. Filling before area removal keeps a nucleus with a dim center from being dropped as small.

Clean the mask first

Before labeling, work on the mask in the Morphological group. Use Fill holesLoading... when uneven staining leaves holes, and Fill small holes when only holes below an area should close. Use Remove small objects (from mask) to drop specks; the reference's example removes every object under 1000 pixels from a DAPI threshold mask. Use Open (erode, dilate) to remove small protrusions, accepting that it reshapes everything.

Then filter the events

After labeling, use the Post Processing group. Remove labels (smaller) and (larger) set a size window in pixels. Remove labels (weaker) and (stronger) set a brightness window, measured in whatever gray image you pass as Image 2, so you can judge DAPI-detected nuclei by a marker channel.

Remove objects (without seeds) keeps the objects a second detection confirms, and Remove objects (with seeds) keeps the ones it missed. The Removal operationsLoading... page shows how to pick among the six.

Layer-level alternatives

The layer's own Post-Processing section has two related engines: Remove objects v2, which keeps events whose chosen measurements fall within a range, and Remove objects touching border, which drops events that reach the sample or region-of-interest border. See Remove ObjectsLoading....

Processing Order

Fill, remove specks, label, delete unwanted events, re-label. Pipeline Ordering, above, gives the reasons.