Statistical Pattern Recognition

Table of contents

• Lab 1. Recognition of a noisy string
• Lab 2. Image segmentation
• Lab 3. Image inpainting
• Lab 4. Interactive foreground extraction

Tasks and mathematical solutions

• solutions.pdf.
• tasks.pdf;

The tasks document is the intellectual property of Valerii Krygin (definability) and is included in this repository with explicit permission.

Acknowledgements

We sincerely thank the course instructor Valerii Krygin (definability) for his dedication, passion, and high-quality support throughout the course.

Setup

To run these applications, you need to have Python3.12.

1. Clone repo

2. Create virtual environment.

   python3.12 -m venv .venv

3. Activate it

   source .venv/bin/activate

4. Install requirements:

   pip install -r requirements.txt

Lab 1 – Recognition of a noisy string

Description

The program converts a string to a noisy image and then decodes it. Dynamic programming algorithm for chain-structured graphical models.

Usage

 $ python3 lab1/decode_string.py --help
usage: decode_string.py [-h] --input_string INPUT_STRING --noise_level NOISE_LEVEL [--seed SEED]

options:
  -h, --help            show this help message and exit
  --input_string INPUT_STRING
                        input string
  --noise_level NOISE_LEVEL
                        noise level of bernoulli distribution
  --seed SEED           seed to debug

Examples

python3 decode_string.py --input_string "billy herrington" --noise_level 0.35 --seed 45

Decoded string: "billy herrington"

Original image	Noisy image	Decoded image

Lab 2 – Image segmentation

Description

The program segmentates a noisy image using Min-Sum Diffusion.

Usage

$ python3 image_denoiser.py --help
usage: image_denoiser.py [-h] --img_path IMG_PATH --alpha ALPHA [--n_iter N_ITER] [--c C [C ...]]

Image segmentation on a noisy image using diffusion.

options:
  -h, --help           show this help message and exit
  --img_path IMG_PATH  Path to the image to denoise
  --alpha ALPHA        Alpha parameter for binary penalties
  --n_iter N_ITER      Number of iterations
  --c C [C ...]        List of colors to segment

Examples

python3 image_denoiser.py --img_path "test_images/map_hsv.png" --alpha 3 --n_iter 100 --c "blue lime"

Noisy image	Segmented image

Usage

python3 image_denoiser.py --img_path "test_images/ipt.png" --alpha 1 --n_iter 100 --c "blue yellow white"

Noisy image	Segmented image

Lab 3 – Image inpainting

Description

The program inpaint mask regions using Tree Reweighted Message Passing (TRW-S) algorithm.

Usage

$ python3 image_inpainter.py --help
usage: image_inpainter.py [-h] --img_path IMG_PATH --alpha ALPHA --epsilon EPSILON --n_labels N_LABELS --n_iter N_ITER

Image inpainter using TRW-S algorithm.

options:
  -h, --help           show this help message and exit
  --img_path IMG_PATH  Path to the image.
  --alpha ALPHA        Smoothing coefficient for binary penalties.
  --epsilon EPSILON    Special parameter, which is responsible for lack of color information.
  --n_labels N_LABELS  Number of labels.
  --n_iter N_ITER      Number of iterations.

Examples

python3 image_inpainter.py --img_path "test_images/mona-lisa-damaged.png" --alpha 0.4 --epsilon 0 --n_labels 64 --n_iter 20

Image with marks	Inpainted image

Lab 4 – Interactive foreground extraction

Description

The program extracts the foreground from an image using the GrabCut algorithm.

Gaussian Mixture Model (GMM) is used to model the foreground and background. TRW-S as an energy minimization algorithm.

Usage

$ python3 extract_foreground.py --help
usage: extract_foreground.py [-h] --img_path IMG_PATH --mask_path MASK_PATH --gamma GAMMA --n_bg N_BG --n_fg N_FG --bg_color BG_COLOR --fg_color FG_COLOR --em_n_iter
                             EM_N_ITER --trws_n_iter TRWS_N_ITER --n_iter N_ITER

Foreground extraction using EM and TRW-S algorithms.

options:
  -h, --help            show this help message and exit
  --img_path IMG_PATH   Path to the input image.
  --mask_path MASK_PATH
                        Path to the interactive mask (user scribbles marking fg/bg).
  --gamma GAMMA         Smoothness weight for pairwise terms. Controls edge preservation. Range: 10-100.
  --n_bg N_BG           Number of Gaussian components for background GMM.
  --n_fg N_FG           Number of Gaussian components for foreground GMM.
  --bg_color BG_COLOR   Color marking background in the mask (e.g., 'red', 'blue').
  --fg_color FG_COLOR   Color marking foreground in the mask (e.g., 'green', 'yellow').
  --em_n_iter EM_N_ITER
                        Number of EM iterations for GMM fitting per iteration.
  --trws_n_iter TRWS_N_ITER
                        Number of TRW-S message passing iterations per iteration.
  --n_iter N_ITER       Total number of refinement iterations (alternating EM and TRW-S).

Examples

python3 extract_foreground.py  \
        --img_path "test_images/alpaca.jpg"  \
        --mask_path "test_images/alpaca-segmentation.png"  \
        --gamma 50  \
        --n_bg 3  \
        --n_fg 3  \
        --bg_color blue  \
        --fg_color red  \
        --em_n_iter 10  \
        --trws_n_iter 10  \
        --n_iter 1 

Image	Manually Marked Mask (Blue - background, Red - foreground)	Segmentation Mask Result	Foreground Result

python3 extract_foreground.py  \
        --img_path "test_images/alpaca.jpg"  \
        --mask_path "test_images/alpaca-segmentation.png"  \
        --gamma 50  \
        --n_bg 3  \
        --n_fg 3  \
        --bg_color blue  \
        --fg_color red  \
        --em_n_iter 10  \
        --trws_n_iter 10  \
        --n_iter 1 

Image	Manually Marked Mask (Green - background, Blue - foreground)	Segmentation Mask Result	Foreground Result

Authors

• Maksym Shylo (@maksymshylo)
• Ruslan Khomenko (@Ruslan3584)