Table of contents

  1. Table of contents
  2. Homework Assignment: Template Matching and Image Analysis
    1. Objective
    2. Tasks
      1. Task 1: Matrix Template Matching
      2. Task 2: Image Template Matching
      3. Task 3: Extensions (Optional, without any grade)

Homework Assignment: Template Matching and Image Analysis

Objective

The objective of this homework assignment is to introduce students to template matching using the sum of squared differences method. Students will apply this technique to matrices and then extend it to grayscale images.

Tasks

Task 1: Matrix Template Matching

  1. Matrix Preparation:
    • Create two matrices, image_matrix and template_matrix, with the following values: 
      image_matrix = np.array([
    [0, 1, 2, 10],
    [3, 4, 5, 11],
    [6, 7, 8, 12]])

template_matrix = np.array([ 
    [4, 5],
    [7, 12]])
    • The dimensions of image_matrix should be larger than template_matrix.
  2. Implement sum_of_squared_differences:
    • Write a Python function called sum_of_squared_differences(image, template) that computes the sum of squared differences between the image and the template.
    • You can use the sum_of_squared_differences function from the Template-Matching.ipynb notebook.
  3. Find the Best Match:
    • Find the location of the best match of template_matrix within the image_matrix. For this porpose, modify the template_matching` function of the mentioned notebook, so that runs without CV2 library.
    • Show the results.

Task 2: Image Template Matching

  1. Image Preparation:
    • Use the coin image from here (or use any other image of your choice).
    • Convert the image to grayscale.
  2. Apply Template Matching:
    • Use the previous functions to find the best match of the coin image within the larger image.
    • Display the original image, the template, and the matched region.

Task 3: Extensions (Optional, without any grade)

  1. Scale and Rotation Invariance:
    • Discuss how template matching can be extended to handle scale and rotation variations.
    • Explore other methods (e.g., normalized cross-correlation) that are more robust to such variations.
  2. Performance Optimization:
    • Investigate ways to optimize the template matching process (e.g., using integral images or FFT-based methods).

  3. Experiment with different templates and images.

  4. Discuss the limitations of template matching (e.g., sensitivity to lighting changes, occlusions, and noise).