Reject background-like regions after segmentation — reject_background

Applies a simple, explicit region-level background classifier to a SAGUI segmentation. This is designed to be used after an intentionally loose support and optional oversegmentation: let the clustering isolate candidate structures, then remove regions with weak multi-band support.

Usage

reject_background_regions(
  segmentation,
  evidence = NULL,
  evidence_count = NULL,
  min_median_evidence = NULL,
  min_median_persistence = 1,
  low_evidence_quantile = 0.2,
  reject_edge_low_evidence = TRUE,
  reject_large_low_evidence = TRUE,
  large_area_quantile = 0.75
)

Arguments

segmentation: A result returned by segment_regions() or segment_regions_large().
evidence: Optional 2-D evidence map. If NULL, the function tries to use segmentation$support$details$evidence.
evidence_count: Optional 2-D band-persistence map. If NULL, the function tries to use segmentation$support$details$evidence_count.
min_median_evidence: Minimum median evidence required for a region. If NULL, uses low_evidence_quantile of region median evidence values.
min_median_persistence: Minimum median band-persistence count required for a region.
low_evidence_quantile: Quantile used to define low-evidence regions when min_median_evidence = NULL.
reject_edge_low_evidence: Logical; reject edge-connected regions with low median evidence.
reject_large_low_evidence: Logical; reject large regions with low median evidence.
large_area_quantile: Quantile of region areas used to identify large regions when reject_large_low_evidence = TRUE.

Value

The input segmentation with cluster_map updated and with background_diagnostics and background_regions appended.