Assignment 1: Colorspaces, Sampling and Filtering

Due April 1st, 2026 at 11:59pm
github details

Overview

This assignment tests your ability to implement core image processing algorithms from scratch. Each of the 6 problems requires you to implement one function covering topics from Lectures 2–4: camera models, colour spaces, interpolation, image pyramids, convolution, and template matching.

Estimated Time: 8–12 hours

Learning Objectives

By completing this assignment, you will:

  • Project 3D world points to 2D image coordinates using the pinhole camera model (Lecture 2)
  • Convert between RGB and HSV colour spaces (Lecture 2)
  • Rotate images using inverse mapping and bilinear interpolation (Lecture 3)
  • Blend images seamlessly using Laplacian pyramid blending (Lecture 3)
  • Implement 2D convolution from scratch with boundary handling (Lecture 4)
  • Implement Normalised Cross-Correlation (NCC) for template matching (Lecture 4)

Getting Started

Prerequisites

  • Python 3.11 or higher
  • Git
  • uv (recommended) — fast Python package manager

Setup

git clone <your-repo-url>
cd 01-filtering-and-edges
make setup   # creates virtual environment and installs dependencies
make test    # run all tests

Accept the assignment on GitHub Classroom using the link above, then clone your personal repository.

Assignment Tasks

Each problem has one function for you to implement. Do not use the restricted libraries listed in the problem descriptions.

Problem 1: 3D Point Projection (Lecture 2)

problem_1_projection.py — Project 3D world points onto a 2D image plane using the pinhole camera model.

Function: project_points(points_3d, K, R, t) → (N, 2) pixel coordinates

Problem 2: RGB to HSV Conversion (Lecture 2)

problem_2_color_spaces.py — Convert an RGB image to HSV colour space from scratch.

Function: rgb_to_hsv(image) → (H, W, 3) array with Hue, Saturation, Value channels

Problem 3: Image Rotation (Lecture 3)

problem_3_rotation.py — Rotate a grayscale image using inverse mapping and bilinear interpolation.

Function: rotate_image(image, angle_degrees) → rotated (H, W) image

Problem 4: Laplacian Pyramid Blending (Lecture 3)

problem_4_pyramid_blending.py — Blend two images seamlessly using Laplacian pyramid blending. Helper functions for Gaussian blur, downsampling, and upsampling are provided.

Function: blend_images(image1, image2, mask, num_levels) → blended (H, W) image

Problem 5: 2D Convolution (Lecture 4)

problem_5_convolution.py — Implement 2D convolution from scratch with boundary handling.

Function: convolve2d(image, kernel, boundary, mode) → convolved image

Do not use scipy.signal.convolve, numpy.convolve, or cv2.filter2D.

Problem 6: Normalised Cross-Correlation (Lecture 4)

problem_6_template_matching.py — Implement NCC template matching from scratch.

Function: compute_ncc(image, template) → NCC score map in $[-1, 1]$

Do not use scipy.signal.correlate, cv2.matchTemplate, skimage.feature.match_template, or numpy.lib.stride_tricks.sliding_window_view.

Grading

30 unit tests are auto-graded via GitHub Actions. Each passed test contributes proportionally toward a maximum score of 100.

Problem Topic Tests Lecture
1. Projection Pinhole camera 3D → 2D 5 2
2. Colour Spaces RGB → HSV conversion 5 2
3. Rotation Inverse mapping + bilinear interpolation 5 3
4. Pyramid Blending Laplacian pyramid blending 5 3
5. Convolution 2D convolution from scratch 5 4
6. Template Matching Normalised Cross-Correlation 5 4
Total 30

Submission

Push your completed code to GitHub. Tests run automatically on each push — check the Actions tab in your repository to see results.

git add .
git commit -m "Complete Assignment 1"
git push origin main

Academic Integrity

Work on this assignment individually. You may consult documentation and tutorials. If you use AI tools (chatbots), acknowledge this in a comment in your code.