Computer Vision Interview
20 essential Q&A
Updated 2026
Harris
Harris Corner Detector: 20 Essential Q&A
Second-moment matrix, response function, and why corners score high.
~11 min read
20 questions
Intermediate
HarrisM matrixeigenvaluesk
Quick Navigation
1
What does the Harris corner detector find?
⚡ easy
Answer: Locations where intensity changes strongly in two directions—corners and strong junctions—via local second-order structure of gradients.
2
What does Harris maximize?
📊 medium
Answer: Change in SSD of a patch under small shifts u,v—approximated by quadratic form involving structure tensor M.
3
Define the second-moment matrix M.
🔥 hard
Answer: M = Σ w(x,y) [Ix² IxIy; IxIy Iy²] over a window—captures local gradient covariance; eigenvectors give principal gradient directions.
4
Interpret eigenvalues λ1, λ2 of M?
📊 medium
Answer: Both small: flat; one large, one small: edge; both large: corner (intensity varies along two orthogonal directions).
5
Harris response R?
📊 medium
Answer: R = det(M) − k·trace(M)² = λ1λ2 − k(λ1+λ2)²—avoids explicit eigen decomposition; k ≈ 0.04–0.06 typical.
6
Effect of k?
⚡ easy
Answer: Tunes sensitivity vs noise; too large suppresses corners; empirical constant, not learned from data in classical form.
7
R on flat region?
⚡ easy
Answer: det≈0, trace≈0 → R negative or near zero—rejected.
8
R on edge?
⚡ easy
Answer: One eigenvalue ~0 → det≈0 while trace>0 → R negative—rejected as corner.
9
What is Shi-Tomasi “good features to track”?
📊 medium
Answer: Score = min(λ1, λ2) with threshold—more stable for tracking; picks corners with minimum directional strength guaranteed.
10
Effect of window size?
📊 medium
Answer: Larger window: smoother M, less localization noise but merges nearby corners; smaller: noisier, better localization.
11
Why Gaussian weights w?
⚡ easy
Answer: Emphasize center of patch, reduce boundary artifacts when sliding window—standard in cornerHarris.
12
Invariances of Harris?
📊 medium
Answer: Invariant to rotation (eigenvalues of symmetric M); not scale invariant—same corner changes type across scales; partial brightness affine in practice.
13
How fix scale weakness?
🔥 hard
Answer: Multi-scale Harris, scale-space extrema (like SIFT), or detectors with inherent scale selection (LoG).
14
Refine corners to sub-pixel?
📊 medium
Answer: Fit quadratic to corner response surface or iterative refinement (OpenCV
cornerSubPix) using gradients.
15
Need NMS?
⚡ easy
Answer: Yes—Harris map is dense; keep local maxima above threshold separated by minimum distance.
16
Harris vs FAST?
📊 medium
Answer: FAST: speed-optimized segment test, not gradient matrix—faster, less accurate localization; Harris more principled, slower.
17
OpenCV
cornerHarris output?
⚡ easy
Answer: Float response map; threshold + NMS to get points; often followed by
goodFeaturesToTrack (Shi-Tomasi).
R = cv2.cornerHarris(gray, 2, 3, 0.04)
18
Why use det − k·trace²?
🔥 hard
Answer: Algebraic proxy for “both eigenvalues large” without sqrt—computationally cheap and continuous score.
19
Gradients Ix, Iy?
⚡ easy
Answer: Usually Sobel or Scharr on smoothed image—noise reduction before derivative recommended.
20
Planar surface assumption?
📊 medium
Answer: Harris assumes small motion model in image plane—breaks for strong perspective on 3D corners unless patch small enough.
Harris Cheat Sheet
M matrix
- Sum of outer products
- Weighted window
λ’s
- Flat / edge / corner
Limits
- Not scale-inv.
- Shi-Tomasi for track
💡 Pro tip: Corner = two strong eigenvalues; edge = one.
Full tutorial track
Go deeper with the matching tutorial chapter and code examples.