Skip to contents

Automated ROI Segmentation via EBImage

Source: R/roi.r
roi_segment_ebimage.Rd

Performs automated background removal using a hybrid pipeline of global thresholding (Otsu's method), morphological operations, and connected component analysis.

Usage

roi_segment_ebimage(
  img_obj,
  method = "otsu",
  keep_largest = TRUE,
  morphology = TRUE
)

Arguments

img_obj

A 'BioThermR' object.

method

String. The thresholding algorithm. Currently, only "otsu" is supported.

keep_largest

Logical. If TRUE (default), keeps only the largest connected object and converts all other clusters to background (NA). This is a powerful denoising step for animal experiments.

morphology

Logical. If TRUE (default), applies morphological opening and closing operations to smooth edges and reduce noise.

Value

A 'BioThermR' object where the processed matrix has been masked (background pixels are set to NA).

Details

The segmentation pipeline consists of four steps:

  1. Normalization: The temperature matrix is scaled to [0, 1] to be compatible with 'EBImage'.

  2. Thresholding: Otsu's method is used to calculate an optimal global threshold that separates the foreground (subject) from the background based on histogram bimodality.

  3. Morphology: A disc-shaped brush (size=5) is used for 'Opening' (to remove salt noise) and 'Closing' (to fill small holes inside the subject).

  4. Component Filter: If keep_largest is TRUE, the function labels all connected regions and retains only the largest one (assuming the animal is the largest heat source), effectively removing smaller artifacts like bedding noise or reflections.

Examples

if (FALSE) { # \dontrun{
# Load raw data
obj <- read_thermal_flir("mouse.jpg")

# Apply automated segmentation
obj <- roi_segment_ebimage(obj, keep_largest = TRUE)
} # }