This repository implements a structured collection of classical computer vision pipelines using MATLAB.
The focus is on deterministic image transformations, multi-resolution analysis, morphology, registration, texture modeling, and feature-based similarity — implemented with explicit parameter control and reproducible export-first design.
This is not a collection of isolated scripts.
It is a modular, systemized exploration of foundational image processing mechanics.
- Deterministic pipelines (no hidden randomness)
- Export-first execution (all artifacts saved)
- Explicit parameterization
- Clear separation of processing stages
- Transparent baselines before optimization
- No black-box abstractions
Each module isolates a specific representation principle in classical vision systems.
RGB / HSV decomposition, grayscale projection, histogram equalization, and distribution remapping.
Channel-wise modeling across RGB, HSV, YIQ with contrast stretching and masking.
Custom convolution, salt-and-pepper noise, Gaussian smoothing, Canny edges.
Fourier decomposition, Gaussian low-pass filtering, spatial-frequency equivalence analysis.
Multi-level wavelet analysis, detail suppression, reconstruction sensitivity.
Sobel edge detection, dilation, hole filling, erosion-based smoothing.
Rotation + translation grid search using correlation maximization.
Texture modeling and similarity ranking using handcrafted descriptors.
Each module ships with an accompanying notes.html (and/or notes.md) that explains:
- the underlying concept (what problem the operation solves),
- the math/intuition (what changes in the representation),
- parameter effects and common failure modes,
- how to interpret exported outputs.
These notes are designed as explorable documentation for reproducible classical vision baselines.
How to use: open notes.html in a browser (or use GitHub Pages if enabled).
Tip: start with Module 01 and follow the modules in order.
- Classical CV focus (no deep learning)
- Multi-representation analysis (spatial, frequency, wavelet)
- Morphological reasoning
- Feature-based similarity
- Explicit complexity trade-offs (e.g., brute-force search)
- Students strengthening foundational CV understanding
- Engineers reviewing classical baselines
- Researchers needing interpretable preprocessing systems
- Anyone bridging classical vision to modern ML
- MATLAB
- Image Processing Toolbox
- Computer Vision Toolbox (for LBP module)
- Wavelet Toolbox (for DWT module)
Modern computer vision is dominated by deep learning.
However, understanding classical image representation, frequency decomposition, morphology, and feature descriptors remains essential for:
- Debugging learned systems
- Designing preprocessing pipelines
- Interpreting model behavior
- Building hybrid classical-ML systems
This repository serves as a structured reference for those foundational principles.
- Each module contains its own README.
- All scripts assume execution from their respective folder.
- Outputs are written to local
output/directories.
Abdul Rafay Mohd
Artificial Intelligence | Medical AI | Computer Vision
This repository is released under the MIT License.