CS 143 / Project 2 / Local Feature Matching

Purpose

For this assignment, we were asked to write three major steps of a local feature matching algorithm: 1) Interest point detection 2) Local feature description and 3) Feature matching.


Approach

Design
I was able to complete the Harris corner detector code, but I had trouble with the SIFT-like local feature descriptor. My approach was to use the eight orientation filter masks to convolve with the input image. I was/am having a hard time with this code in trying to perform post-processing once the image patches were convolved. Summing up the convolved patches for the current image space could be a possible solution.



Results

Harris corner detector works and returns quite a few possible interest points depending on the threshold (lower = more possible interest points). The code I was writing for the SIFT-like local feature descriptor has been commented out and does not work at this point. Without the SIFT-like local feature descriptor, the feature matching is irrelevant.

Lessons Learned

1) Start early especially when multiple projects are due at the same time. 2) Utilize the TAs and Professor Hays as often as needed (during office hours and within reason). 3) Manage my time better.