Image Processing — Unit 2

High Performance Computing | Universidad Politecnica de Yucatan (UPY) Data Engineering — 7th Quadrimester | March 2026 Students:

• Balam Euan Cesar Arturo
• De Gante Pérez Diego
• Polanco González Roselyn

---

Overview

Implementation and performance comparison of three classical grayscale image filters under three computational strategies:

Strategy	Description
Pure Python	Standard Python loops; no external libraries in filter logic. Serves as the performance baseline.
NumPy	Vectorised array operations delegated to NumPy's C/BLAS back-end.
Cython	Compiled native extension with typed memoryviews and C-level arithmetic.

---

Filters Implemented

Filter	Type	Purpose
Gaussian	Linear	Weighted-average blur for noise reduction
Sobel	Linear	Gradient-magnitude edge detection (Kx + Ky)
Median	Non-linear	Impulse (salt-and-pepper) noise removal

---

Project Structure

EA2_U2/
│
├── src/                          # All filter source code
│   ├── filters_pure_python.py    # Pure Python implementations (baseline)
│   ├── filters_numpy.py          # NumPy vectorised implementations
│   ├── filters_cython.pyx        # Cython source (compiled to native extension)
│   └── setup.py                  # Build script for the Cython extension
│
├── scripts/
│   └── main.py                   # Benchmark runner + visualisation + summary table
│
├── assets/
│   └── test_image.png            # Input image (auto-generated if missing)
│
├── output/                       # All generated images (auto-created on first run)
│   ├── out_pure_gaussian.png
│   ├── out_pure_sobel.png
│   ├── out_pure_median.png
│   ├── out_numpy_gaussian.png
│   ├── out_numpy_sobel.png
│   ├── out_numpy_median.png
│   ├── out_cython_gaussian.png
│   ├── out_cython_sobel.png
│   ├── out_cython_median.png
│   ├── comparison_grid.png
│   └── timing_chart.png
│
├── report/
│   └── Image_Processing_Unit2_Report.docx
│
├── .gitignore
├── requirements.txt
└── README.md

Note: assets/ and output/ are created automatically on the first run if they do not exist. You do not need to create them manually.

---

Requirements

• Python >= 3.10
• A C compiler (GCC / Clang on Linux or macOS; MSVC or MinGW on Windows)

---

Setup

Step 1 — Install Python dependencies

From the project root:

pip install -r requirements.txt

Step 2 — (Optional) Build the Cython extension

The Pure Python and NumPy filters run without any build step. Cython requires compilation:

cd src
python setup.py build_ext --inplace
cd ..

This produces a platform-specific shared library inside src/, for example:

src/filters_cython.cpython-311-x86_64-linux-gnu.so   (Linux)
src/filters_cython.cpython-311-darwin.so              (macOS)
src/filters_cython.cp311-win_amd64.pyd                (Windows)

If the extension is not built, main.py will skip the Cython results and print a reminder, but the rest of the benchmark runs normally.

---

Running

From the project root (EA2_U2/):

# Use the auto-generated synthetic test image
python scripts/main.py

# Use your own image
python scripts/main.py path/to/your/image.png

All output images and charts are saved to output/ automatically.

---

Output Files

File	Description
output/out_pure_gaussian.png	Gaussian — Pure Python
output/out_pure_sobel.png	Sobel — Pure Python
output/out_pure_median.png	Median — Pure Python
output/out_numpy_gaussian.png	Gaussian — NumPy
output/out_numpy_sobel.png	Sobel — NumPy
output/out_numpy_median.png	Median — NumPy
output/out_cython_gaussian.png	Gaussian — Cython
output/out_cython_sobel.png	Sobel — Cython
output/out_cython_median.png	Median — Cython
output/comparison_grid.png	Side-by-side filter comparison (NumPy)
output/timing_chart.png	Grouped bar chart of execution times (log scale)

---

Benchmark Results (256 x 256 image)

Filter	Pure Python (s)	NumPy (s)	Cython (est., s)	Speedup (Py / Cy)
Gaussian	0.5992	0.0025	~0.0003	~2000x
Sobel	0.2059	0.0040	~0.0006	~340x
Median	0.0923	0.0258	~0.0038	~24x

---

.gitignore

Recommended entries to keep the repository clean:

# Cython build artifacts
src/*.so
src/*.pyd
src/*.c
src/*.html
build/

# Generated output images (large binary files)
output/

# Python cache
__pycache__/
*.pyc