Global Black-and-White Thresholding¶

Applies a histogram-based thresholding algorithm to convert a grayscale image into a binary (black-and-white) result and writes it to the Selection or Mask layer.

Overview¶

Accessible via Ribbon → Tools → Semi-automatic segmentation → Global thresholding.

Twelve algorithms are available, each suited to different image characteristics. The result is written to the Selection or Mask layer depending on the Destination setting.

Global black and white thresholding in MIB

Controls¶

• Algorithm: thresholding method to use (see Algorithms below for descriptions of each).

Options panel

• Mode: dataset extent to process — 2D, Slice (current slice only), 3D, Stack (all slices in the current Z-stack), or 4D, Dataset (all slices and time points).
• Color channel: color channel used for thresholding.
• Destination: layer that receives the binary result — selection or mask.
• Fraction of foreground pixels + slider: fraction of pixels assumed to belong to the foreground (0–1). Active only for the Percentile algorithm.
• Threshold offset + slider: integer offset added to the computed threshold — positive raises it (fewer foreground pixels), negative lowers it (more foreground pixels).

• Reset sliders: resets Fraction of foreground pixels to 0.5 and Threshold offset to 0.

• auto preview: updates the Selection layer automatically after each widget change (current slice only).

• Preview: applies thresholding to the current slice and shows the result in the Selection layer without committing to the chosen scope.
• Apply: applies thresholding to the full selected scope.
• Close: closes the dialog.

Algorithms¶

Concavity¶

Suitable when images lack distinct objects and background, making MINIMUM and INTERMODES algorithms ineffective:

• Constructs the convex hull H of the histogram y
• Finds local maxima of |H − y|
• Sets threshold t to the value of j maximising the balance measure bj = Aj(An − Aj)

Entropy¶

A maximum entropy method that splits the histogram into two probability distributions (objects / background):

• Sets threshold t to maximise the sum of entropies of those distributions:
  (Ej / Aj) − log(Aj) + ((En − Ej) / (An − Aj)) − log(An − Aj)

InterMeans iter¶

An iterative algorithm (similar to Otsu but less computationally intensive):

• Starts with an initial threshold t; iteratively updates t = ⌊(μt + νt) / 2⌋ until convergence
• Results may depend on the initial t value
• Use Mean for comparable object and background areas; InterModes for small objects

InterModes¶

Assumes a bimodal histogram — an alternative to Minimum:

• Identifies two peaks yj and yk and sets t = (j + k) / 2
• Unsuitable for histograms with extremely unequal peaks

Mean¶

Sets t to the integer part of the mean of all pixel values: t = Bn / An.

Does not take histogram shape into account — often yields suboptimal results.

Median¶

Sets t to the median of all pixel values (equivalent to Percentile with Fraction = 0.5).

Does not take histogram shape into account.

MinError¶

Assumes a Gaussian mixture model and minimises classification error:

• Allows different means and variances for objects and background
• Sets t to minimise pj log(σj / pj) + qj log(τj / qj)

MinError iter¶

Iterative version of MinError — less computationally intensive:

• Initialises t using Mean and iterates until convergence
• Fails if the resulting quadratic equation has no real solution

Minimum¶

Assumes a bimodal histogram:

• Smooths the histogram with a three-point mean filter until only two local maxima remain
• Chooses t at the valley between them (yt−1 > yt < yt+1)
• Unsuitable for histograms with unequal peaks or a broad, flat valley

Moments¶

Moment-preserving thresholding — sets t such that the first three moments of the gray-level image are preserved in the thresholded binary result.

Otsu¶

Implemented via MATLAB's graythresh. Calculates an optimal threshold to minimise intra-class intensity variance. Works with the full bit depth of the image.

Percentile¶

Assumes that the fraction of foreground pixels is known:

• Sets t to the highest gray-level such that at least (1 − Fraction) of pixels map to background
• Configure via Fraction of foreground pixels (default 0.5 = Median)

Batch scripting¶

This tool supports batch scripting for automation.

BatchOpt.Algorithm    = {'Otsu'};        % thresholding algorithm
BatchOpt.Mode         = {'3D, Stack'};   % '2D, Slice' | '3D, Stack' | '4D, Dataset'
BatchOpt.ColorChannel = {'Ch 1'};        % color channel
BatchOpt.Destination  = {'selection'};   % 'selection' | 'mask'
% Optional range limiters:
BatchOpt.t = [1 1];      % time points [t1, t2]
BatchOpt.z = [10 20];    % slices [z1, z2]
BatchOpt.x = [10 120];   % x crop [x1, x2]
% Percentile-specific:
BatchOpt.ForegroundFraction = {0.3, [0 1], 'off'};
% Threshold fine-tuning:
BatchOpt.ThresholdOffset = {5, [-Inf Inf], 'on'};

obj.mibController.startController('controllers.GlobalThresholding', [], BatchOpt);

Back to MIB | User interface | Ribbon | Tools