Small Scale Multi-Object Segmentation in Mid-Infrared Image Using the Image Timing Features–Gaussian Mixture Model and Convolutional-UNet

. 2025 May 30;25(11):3440. doi: 10.3390/s25113440

Algorithm 1: ITF-GMM

1. Input
Video frame sequence

2. Output
Foreground masks for each frame

3. Initialization
for each pixel x do
Initialize N Gaussian distributions
end for

4. Procedure process
for each frame do
      Extract ROI
      Compute dynamic LR via Equation (11) in ROI
      Compute dynamic Gaussian components via Equations (12)–(13) in ROI
      for each pixel x_t do
            Compare x_t with existing model via Equation (4)
            if a match is found then
                  Mark x_t as background
                  Update parameter via Equations (5)–(7)
            else if no match then
                  Mark x_t as foreground
                  Update weights via Equation (8)
            end if
            if x_t ∈ ROI then
                  update LR and N
            else
                  N = 2
            end if
            Normalize all weights to ensure sum is 1 and N satisfies Equation (9)
      end for
end for