Feature Extraction

Module: Computer Vision | Difficulty: Intermediate

Classical Feature Detectors

Harris Corner Detector

The corner response function:

where

SIFT Features

Scale-Invariant Feature Transform detects keypoints at multiple scales:

1. Build scale-space:
2. Detect extrema in Difference of Gaussians
3. Assign orientation based on gradient histogram
4. Compute 128-dimensional descriptor

ORB (Oriented FAST and Rotated BRIEF)

Efficient alternative combining FAST keypoint detector with BRIEF descriptor, adding rotation invariance.

Feature Matching with BFMatcher

import cv2
import numpy as np

def match_features(img1, img2, method='orb'):
    if method == 'orb':
        detector = cv2.ORB_create(nfeatures=1000)
    elif method == 'sift':
        detector = cv2.SIFT_create()
    
    kp1, des1 = detector.detectAndCompute(img1, None)
    kp2, des2 = detector.detectAndCompute(img2, None)
    
    bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
    matches = bf.match(des1, des2)
    matches = sorted(matches, key=lambda x: x.distance)
    
    return kp1, kp2, matches

def compute_homography(kp1, kp2, matches):
    pts1 = np.float32([kp1[m.queryIdx].pt for m in matches])
    pts2 = np.float32([kp2[m.trainIdx].pt for m in matches])
    H, mask = cv2.findHomography(pts1, pts2, cv2.RANSAC, 5.0)
    return H, mask

Key Takeaways

• SIFT provides scale and rotation invariance
• ORB offers real-time performance with competitive accuracy
• Deep features (from CNNs) now outperform hand-crafted descriptors