Segment compact SED clumps and the diffuse galaxy body

Experimental clump-aware segmentation mode. The method builds a loose foreground support, detects compact SED-anomalous seeds (or accepts supplied clump seeds), grows them into clump footprints, splits each clump into internal profile levels, and runs segment_regions_large() on the remaining diffuse body. The final cluster_map merges clump-profile regions and the diffuse-body segmentation.

Usage

segment_clumps(
  input,
  Ncomp_body = 80,
  bands = NULL,
  support = NULL,
  support_info = NULL,
  support_args = list(),
  clump_search_mask = NULL,
  clump_reject_mask = NULL,
  clump_reject_fraction = 0.5,
  clump_seeds = NULL,
  n_clumps = NULL,
  max_clumps = 50L,
  score_quantile = 0.995,
  min_seed_distance = 3,
  clump_levels = 3L,
  grow_radius = 3L,
  max_radius = 7L,
  min_pixels = 9L,
  footprint_quantile = 0.1,
  core_drop_frac = 0.2,
  sigma_threshold = 3,
  footprint_mode = c("connected", "radial"),
  connectivity = 8L,
  contrast_weight = 0.5,
  sed_weight = 0.5,
  small_sigma = 0.7,
  large_sigma = 4,
  apply_quality_filter = FALSE,
  min_peak_score = 0.35,
  min_median_score = NULL,
  min_peak_contrast = 0.15,
  min_peak_sed_anomaly = 0.15,
  high_score_threshold = 0.35,
  min_high_score_pixels = 3L,
  probable_peak_score = 0.15,
  probable_peak_contrast = 0.15,
  probable_median_contrast = 0.02,
  probable_peak_sed_anomaly = 0.1,
  clump_relax = 0.5,
  body_segment = TRUE,
  knn_k = 40,
  auto_k = FALSE,
  max_k = NULL,
  feature_scale = c("none", "robust_col"),
  spatial_weight = 0.1,
  cluster_pretransform = "none",
  verbose = TRUE
)

Arguments

input: 3-D array or FITS-like list with imDat. This should usually be the PSF-matched cube used for segmentation.
Ncomp_body: Number of diffuse-body regions.
bands: Bands used for support, clump scoring, and body segmentation.
support: Optional precomputed logical support mask.
support_info: Optional support object, e.g. from build_adaptive_support(). If NULL, adaptive support is built.
support_args: Named list passed to build_adaptive_support() when support_info = NULL and support = NULL.
clump_search_mask: Optional logical mask with the same spatial dimensions as the cube. When supplied, automatic clump seeds are searched only inside this mask, while the diffuse body can still be segmented over the full foreground support. Use this to target context-specific compact structures, e.g. off-disc jellyfish knots, in-disc HII regions, or halo compact-source candidates.
clump_reject_mask: Optional logical mask with the same spatial dimensions as the cube. Candidate clump footprints whose overlap with this mask exceeds clump_reject_fraction are removed before profile splitting. This is a generic context filter, not a science label.
clump_reject_fraction: Maximum allowed footprint overlap with clump_reject_mask.
clump_seeds: Optional data frame with clump_id, row, and col. If NULL, seeds are detected automatically.
n_clumps: Optional exact number of automatic clump seeds.
max_clumps: Maximum number of automatic clump seeds.
score_quantile: Quantile threshold for automatic clump seeds.
min_seed_distance: Minimum seed separation in pixels.
clump_levels: Number of internal profile regions per clump.
grow_radius: Initial clump-footprint growth radius.
max_radius: Maximum clump-footprint growth radius.
min_pixels: Minimum pixels per grown clump footprint.
footprint_quantile: Local score quantile used to trim grown footprints.
core_drop_frac: For connected footprints, keep connected pixels above this fraction of the seed-core brightness contrast relative to the local background.
sigma_threshold: For connected footprints, keep connected pixels above this robust local-background significance threshold. Use NULL to disable.
footprint_mode: Footprint-growth mode. "connected" uses the seed only as an anchor and returns the connected high-score component, allowing elongated or irregular clump shapes. "radial" keeps the older compact radius-based growth.
connectivity: Pixel connectivity used by footprint_mode = "connected".
contrast_weight: Weight of compact local contrast in the clump score.
sed_weight: Weight of local SED anomaly in the clump score.
small_sigma: Compact smoothing scale for the clump score.
large_sigma: Background smoothing scale for the clump score.
apply_quality_filter: Logical; if TRUE, discard candidate clumps that fail the evidence cuts before profile splitting.
min_peak_score, min_median_score, min_peak_contrast, min_peak_sed_anomaly: Evidence cuts passed to clump_evidence_table().
high_score_threshold, min_high_score_pixels: High-score-pixel criterion passed to clump_evidence_table().
probable_peak_score, probable_peak_contrast, probable_median_contrast, probable_peak_sed_anomaly: Looser evidence thresholds used for "probable" clumps.
clump_relax: Single global permissiveness factor for the clump evidence cuts, bounded to [0, 1]. 0 is conservative, 0.5 keeps the nominal thresholds, and 1 accepts fainter/weaker compact candidates. This does not use SED-fitting results.
body_segment: Logical; segment the remaining diffuse body.
knn_k, auto_k, max_k, feature_scale, spatial_weight: Sparse-Ward arguments passed to segment_regions_large() for the diffuse body.
cluster_pretransform: Spectral pretransform for diffuse-body segmentation.
verbose: Logical; print progress messages.

Value

A Sagui-like list with merged cluster_map, clump diagnostics, and body segmentation products.