Image analysis

Description

This project solves various problems relating to image analysis, as described in ImageAnalysis.pdf. The analysis report can be found in report/.

Topics:

• Image segmentation
• Inverse problems
• Compressed-sensing
• Wavelet compression
• Optimization
• CT reconstruction (TV-regularization, learned gradient descent)

Project structure

├── report/    # contains report
├── data/      # contains provided data
├── figures/   # directory for storing plots used in the report
├── scripts/   # directory containing all scripts for reproducing results in report
├── src/       # package containing re-usable components used in the scripts
|
├── .gitignore         # specifies untracked files to ignore
├── ImageAnalysis.pdf  # coursework assignment
├── Dockerfile         # containerisation instructions
├── LICENSE            # license for project
├── README.md          # this file
├── environment.yml    # environment specifications
├── env_light.yml      # lightweight environment (not including odl)
├── env_very_light.yml # environment not including odl, pywavelets OR version specs

Usuage

Re-creating environment

Attempt to re-create the full environment for the project:

# Try re-create exact conda environment
$ conda env create -f environment.yml -n <env-name>

If this fails due to conflicts relating to the odl package, you will have to re-create the smaller environment (not containg odl), and the CT reconstruction part of the project can be done using CT_reconstruction.ipynb in google colab (More details below). The smaller environment can be re-created using one of the following:

# Option 1: Conda
$ conda env create -f env_light.yml -n <env-name>

# If env_light didnt work, try:
$ conda env create -f env_very_light.yml -n <env-name>
$ conda activate <env-name>
$ pip install pywavelets

# Option 2: Generate docker image and run container
$ docker build -t <image_name> .
$ docker run -ti <image_name>

Reproducing all results EXCEPT problem 3.2

To re-produce all results/figures presented in the report, use the following commands

# 1.1 CT segmentation
$ python ./scripts/ct_segmentation.py --img ./data/CT.png --output_dir ./figures

# 1.2 Flowers segmentation
$ python scripts/flower_segmentation.py --img ./data/noisy_flower.jpg --output_dir ./figures

# 1.3 Coins segementation
$ python scripts/coins_segmentation.py --img ./data/coins.png --output_dir ./figures

# 2.1 Line fitting
$ python scripts/line_fitting.py --y_line ./data/y_line.txt --y_outlier_line ./data/y_outlier_line.txt --output ./figures/line_fits.png

# 2.2 Compressed sensing
$ python scripts/compressed_sensing.py

# 2.3 Wavelet compression
$ python ./scripts/wavelet_compression.py --img ./data/river_side.jpeg --output_dir ./figures

# 3.1 Gradient descent
$ python ./scripts/gradient_descent.py

In each case, plots will be saved to --output_dir and other results will be printed to the terminal.

Reproducing problem 3.2

If you were able to reproduce the full environment from environment.yml, then run the script below. Also try running the script if you happen to have an environment with the relevant packages installed.

$ python scripts/ct_reconstruction.py --output_dir ./figures

Otherwise, use the notebook CT_reconstruction.ipynb in google colab.

Timing

I ran all scripts on my personal laptop with the following specifications:

• Chip: Apple M1 Pro
• Total Number of Cores: 8 (6 performance and 2 efficiency)
• Memory (RAM): 16 GB
• Operating System: macOS Sonoma v14.0

The ct_reconstruction.py script took ~20 minutes to run. All other scripts ran in less than 2 minutes.