Convolution for Image Filtering
Convolution for Image Filtering
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: THEend8_
Task 1. (50 marks) 2D Convolution for Image Filtering
OpenCV provides a function cv.filter2D() to convolve a kernel with an image (See Lab 03 -
Image Manipulation and Lecture 5.). Given a kernel (or a filter) with any size, do the
following:
1. (25 marks) Reimplement the 2D convolution without using the built-in cv.filter2D()
function.
2. (10 marks) Test and visualize your implementation results.
3. (15 marks) Discuss the difference between your implementation, your results
compared with the OpenCV implementation.
Note:
- You can look at the cv.filter2D() code to understand how OpenCV implements this
function. However, it is mandatory that you reimplement the function based on your
understanding.
Task 2. (50 marks) Feature Extraction
In Lecture 11 and Lab 04 - SIFT & Feature Matching, we have discussed the SIFT feature.
In practice, there are several other feature extraction methods such as SURF or ORB. In this
task, we will do extra reading, implementation, and compare SIFT vs. SURF vs. ORB.
Papers to read:
- Bay et al., SURF: Speeded Up Robust Features, ECCV 2006
- Rublee et al., ORB: An efficient alternative to SIFT or SURF, ICCV 2011.
1. (20 marks) Read the SURF and ORB papers and tutorials, summarize your
understanding. Compare the differences between SIFT vs. SURF vs. ORB.
2. (10 marks) Given two images (victoria.jpg and victoria2.jpg – both available on
Canvas), call OpenCV functions to extract the SURF and ORB keypoints. You can
use the built-in functions from OpenCV. Visualize the detected keypoints.
3. (20 marks) Given two images (victoria.jpg and victoria2.jpg), extract the descriptors
using SIFT, SURF and ORB. Perform keypoint matching using Brute-Force Matcher.
From the results, which method do you think perform the best? Justify your answer.