Histograms & Feature Intro MCQ
Histogram equalization, CLAHE, and an introduction to keypoints, descriptors, and matching.
Histogram Equalization MCQ
Histogram-based enhancement
Histogram equalization spreads intensity usage to improve contrast by remapping via the cumulative distribution—global methods can over-amplify noise in flat regions.
CLAHE
Contrast Limited Adaptive HE applies equalization in tiles and clips histogram peaks to reduce noise amplification and blocking.
Building blocks
CDF
Running sum of normalized histogram defines a monotonic gray-level transform.
Linear stretch
Maps min–max to full range—simple baseline before heavier transforms.
Artifacts
Equalization can wash out natural appearance; clip limit and tile grid matter.
Color images
Apply on luminance channel to avoid hue shifts, or use careful per-channel strategies.
Transform view
Input intensity → T(r) → Output intensity
Feature Detection Intro MCQ
Local features
Many pipelines detect repeatable interest points, compute appearance descriptors, match across views, then estimate geometry (e.g., homography, essential matrix) with RANSAC.
Detector vs descriptor
Detectors propose (x,y,scale,orientation); descriptors encode local patch appearance for discrimination.
Vocabulary
Corners & blobs
Corners have two-direction intensity change; blobs respond to maxima of scale-space filters.
Distance metrics
L2 for float descriptors; Hamming for binary (ORB/BRIEF).
Lowe’s ratio
Reject ambiguous matches when nearest vs second-nearest distance ratio is high.
RANSAC
Fits a model while rejecting outliers in putative correspondences.
Typical flow
Detect → Describe → Match → Geometric verify