Project 2: Feature Detection and Matching

CS5670 - Intro to Computer Vision

Project Details Page: https://www.cs.cornell.edu/courses/cs5670/2023sp/projects/pa2/

Introduction:

The goal of feature detection and matching is to identify a pairing between a point in one image and a corresponding point in another image. These correspondences can then be used to stitch multiple images together into a panorama. In this project, we detected image features and matching pairing features as following:

• Feature detection using Harris
• Feature description (simple and MOPS)
• Feature descriptor SIFT (Scale Invariant Feature Transform) implemented in bonus directory in 'extracredit.py'
• Feature matching (SSD and ratio)

Results:

Input:

Yosemite 1 & Yosemite 2

Output: Keypoint Detection:

Yosemite 1: 1559 keypoints detected

Yosemite 2: 1356 keypoints detected

Output: Feature Matching

MOPS SSD: 1559 Matches

MOPS Ratio: 1559 Matches

Simple SSD: 1559 Matches

Simple Ratio: 1559 Matches

Performance Benchmark

ROC Curves and their respective AUC values for all matcher types and tests are shown below in the figure:

Extra Credit: SIFT ROC Curves

ROC Curves:

SIFT SSD Test ROC Curve: AUC Value 0.9841

SIFT Ratio Test ROC Curve: AUC Value 0.9858

Feature Matching Results:

SIFT SSD

SIFT Ratio