Goal: Learn to implement feature matching of several local feature descriptors

• Understand two keypoint matching methods: template matching and feature descriptor matching.
• Implement the OpenCV feature matching method for 5 keypoint descriptors: SIFT, SURF, ORB, BRISK, FREAK.

Readings

• R. Laganière, OpenCV 3 computer vision application programming cookbook. 3rd, Packt Publishing, 2017.
  • Chapter 9 Describing and Matching Interest Points

Program and test images

• Search the sample code of "Chapter 9 Describing and Matching Interest Points" by yourself.
• The tests images of the book chapter, Notre-Dame Cathedral, are provided for you to download: image1, image2.
• Write programs that can read test images and run the two feature matching methods by OpenCV.
• You have to test your program by your images. You should take photos of an object with different scales and viewpoints.
• Upload your source code (*.c, *.cpp, *.h, *.hpp) and image files by this link.

Report

• Create a web page with descriptions, explanation and pictures for your programs. Requirements of the report page:
  • For each program code, you have to write 4 parts: (1) goal of this code, (2) theory and principle of the code, (3) code segment explanation, and (4) result comparison or analysis.
• Theory explanation
  • (1) template matching,
  • (2) knn matching,
  • (3) radius matching,
  • (4) cross check,
  • (5) ratio test.
• Code explanation
  • You have to explain at least two important OpenCV methods: cv::matchTemplate function and cv::BFMatcher class. Note that the cv::BFMatcher class has three match functions: match(), knnMatch() and radiusMatch().
  • You have to explain the most important code segments in your program.
• Experiments
  • Use the Notre-Dame images and your images to run your programs.
  • Change parameters of algorithm's functions to get different result images.
  • Compare and discuss the result images, and explain why the change of parameters can produce different results.