Classical Image Processing Algorithms

Manual implementations of core computer vision routines: FFT magnitude/phase reconstruction, spatial vs frequency-domain filtering, stereo disparity via normalized cross-correlation, and a Canny edge detector pipeline.

Highlights

• Reconstructs images from swapped FFT magnitude and phase components.
• Compares spatial convolution with frequency-domain filtering.
• Computes stereo disparity maps with manual NCC and compares against OpenCV StereoBM.
• Implements Gaussian smoothing, gradients, non-maximum suppression, double thresholding, and hysteresis for Canny-style edge detection.

Repository Layout

• fft_phase_reconstruction.py - Fourier magnitude/phase experiments.
• spatial_frequency_filtering.py - manual convolution and FFT-based filtering.
• stereo_ncc_disparity.py - stereo matching with normalized cross-correlation.
• canny_from_scratch.py - edge detection pipeline from first principles.
• data/ - sample images and generated comparison outputs.

Setup

pip install -r requirements.txt

Run

python fft_phase_reconstruction.py
python spatial_frequency_filtering.py
python stereo_ncc_disparity.py
python canny_from_scratch.py

Algorithm output

Representative outputs from manual Canny, stereo NCC, and frequency-domain filtering.

Image-processing scope

• Core image-processing algorithms implemented directly rather than hidden behind one-line library calls.
• Comparison artifacts that make intermediate behavior visible.
• Breadth across Fourier filtering, stereo matching, and edge detection.

Follow-up validation

• The implementations prioritize clarity over real-time performance.
• Parameter choices are tuned for the included sample images.
• Next steps: add benchmark timings and a small pytest suite for numerical regression checks.