🔑 Keypoints & Descriptors (SIFT, SURF, ORB)

Discover how to locate distinctive points in images (keypoints), describe their local neighbourhoods (descriptors) and compare methods in terms of speed, invariance and licence.

---

What Are Keypoints?

Keypoints are visually distinctive and stable points in an image—often corners, blobs or edges. Each keypoint encodes:

• Position: ((x, y)) coordinates
• Scale: size of the feature
• Orientation: angle of the feature

---

✅ Summary

Method	Descriptor	Best For
SIFT	128 floats	Accuracy, scale invariance
SURF	64 floats	Speed + accuracy (licensed)
ORB	32 binary	Fast, open-source applications

What Are Descriptors?

Descriptors are numerical vectors (e.g. 32–128 dimensions) that describe the local neighbourhood around a keypoint. Think of them as “fingerprints” you can compare across images to match keypoints.

---

Key Feature Detectors & Descriptors

Method	Type	Speed	Patent	Description
SIFT	Keypoints + Descriptors	Slow	Patented (pre-2020)	Robust to scale/rotation, precise
SURF	Faster SIFT	Fast	Patented	Scale-invariant, faster but licensed
ORB	Open-source + Fast	Fast	Free	Efficient alternative to SIFT/SURF

---

🐍 ORB Example

import cv2
import matplotlib.pyplot as plt

# Load image in grayscale
img = cv2.imread("bird1.jpg", cv2.IMREAD_GRAYSCALE)

# Create ORB detector
orb = cv2.ORB_create()

# Detect keypoints and compute descriptors
keypoints, descriptors = orb.detectAndCompute(img, None)

# Draw keypoints
img_orb = cv2.drawKeypoints(img, keypoints, None, color=(0, 255, 0))

# Display
plt.imshow(img_orb, cmap='gray')
plt.title(f"ORB Keypoints: {len(keypoints)} found")
plt.axis("off")
plt.show()

🐍 SIFT Example

import cv2
import matplotlib.pyplot as plt

# Load image in grayscale
img = cv2.imread("bird1.jpg", cv2.IMREAD_GRAYSCALE)

# Create SIFT detector (requires opencv-contrib-python)
sift = cv2.SIFT_create()

# Detect keypoints and compute descriptors
keypoints, descriptors = sift.detectAndCompute(img, None)

# Draw rich keypoints
img_sift = cv2.drawKeypoints(
    img, keypoints, None,
    flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS
)

# Display
plt.imshow(img_sift, cmap='gray')
plt.title(f"SIFT Keypoints: {len(keypoints)} found")
plt.axis("off")
plt.show()

Note: SIFT is available via opencv-contrib-python.

Descriptor Details

Feature	ORB	SIFT
Descriptor Size	32 or 64 dimensions	128 floats
Data Type	Binary	Float32
Matching Speed	Very fast	Slower
Rotation Invariance	✅	✅
Scale Invariance	❌ (partial)	✅