(SEM V) THEORY EXAMINATION 2024-25 COMPUTER VISION

Subject Code: BCIT053
Maximum Marks: 70
Time: 3 Hours
Paper ID: 310322

Question Paper Overview

SECTION A (2 × 7 = 14 Marks)

(Short answer questions — key foundational concepts in Computer Vision)

a. Define edge detection in image processing.
b. What is the Sobel operator used for?
c. What is object labeling?
d. What are centroidal profiles?
e. What is the purpose of line localization in the Hough Transform (HT)?
f. Explain optical flow in motion analysis.
g. What is the significance of multiple camera views in surveillance?

SECTION B (Attempt any three × 7 = 21 Marks)

(Descriptive or applied theory questions)

a. What is mathematical morphology? Explain its basic operations with examples.
b. Explain the concept of skeletonization and thinning in image processing.
c. What is ellipse detection using Hough Transform (HT)? Explain the process in detail.
d. Explain the projection schemes used in 3D vision.
e. Discuss the application of active appearance models in face recognition.

SECTION C (Attempt one part from each question × 7 = 35 Marks)

Q3

(a) What is the significance of texture descriptors in image classification?
OR
(b) What is watershed segmentation? Explain its relevance in texture analysis.

Q4

(a) What are Fourier descriptors? How are they used in image processing?
OR
(b) Discuss the applications of centroidal profiles in medical imaging.

Q5

(a) Explain human iris location as a case study of the Hough Transform.
OR
(b) What is the significance of the foot-of-normal method in straight-line detection?

Q6

(a) What is layered motion? Discuss its significance in visual analysis.
OR
(b) What are the applications of bundle adjustment in aerial imaging?

Q7

(a) What are the methods for detecting road markings in vehicle vision systems?
OR
(b) Explain the process of human gait analysis in biometric identification.

Key Topics for Revision

1. Edge Detection

Process of identifying points where intensity changes sharply.

Used to detect object boundaries.

Common operators: Sobel, Prewitt, Canny, Laplacian of Gaussian.

2. Sobel Operator

A gradient-based filter that detects edges by approximating first derivatives of image intensity.

Uses two kernels for horizontal & vertical gradients:

• Gx=[−10+1−20+2−10+1],Gy=[−1−2−1000+1+2+1]G_x = \begin{bmatrix} -1 & 0 & +1 \\ -2 & 0 & +2 \\ -1 & 0 & +1 \end{bmatrix}, \quad G_y = \begin{bmatrix} -1 & -2 & -1 \\ 0 & 0 & 0 \\ +1 & +2 & +1 \end{bmatrix}Gx​=​−1−2−1​000​+1+2+1​​,Gy​=​−10+1​−20+2​−10+1​​

3. Object Labeling

Process of assigning unique labels to connected components in a binary image.

Example: Label all white pixels connected together as a single object.

Techniques: 4-connectivity and 8-connectivity.

4. Mathematical Morphology

Based on set theory and used for shape analysis.

Basic operations:

Erosion: Removes small white noise.

Dilation: Expands object boundaries.

Opening: Erosion followed by dilation (removes noise).

Closing: Dilation followed by erosion (fills gaps).

5. Skeletonization and Thinning

Reduce binary objects to their 1-pixel-wide skeletons.

Used in shape analysis, OCR, and feature extraction.

6. Hough Transform (HT)

Detects parametric shapes (lines, circles, ellipses).

Each edge point votes in a parameter space (accumulator).

Line localization: Identifies lines using ρ–θ representation.

Ellipse detection: Extended HT with parameters (x₀, y₀, a, b, θ).

7. Texture Descriptors

Quantify surface patterns in images.

Examples:

Local Binary Pattern (LBP)

Gray-Level Co-occurrence Matrix (GLCM)

Gabor filters

Used in image classification, medical imaging, and remote sensing.

8. Fourier Descriptors

Represent object boundary shapes using Fourier coefficients.

Steps:

Extract boundary coordinates.

Apply Discrete Fourier Transform (DFT).

Use coefficients as shape features (invariant to translation/rotation).

9. Centroidal Profiles

Plot of distance from object centroid to boundary points as a function of angle.

Used in shape matching, medical imaging (organ or tumor boundary analysis).

10. Optical Flow

Vector field representing apparent motion of objects in sequential frames.

Methods: Horn–Schunck, Lucas–Kanade.

Used in motion tracking, navigation, and video stabilization.

11. Layered Motion

Scene is divided into motion layers (background, objects).

Each layer modeled with its own motion parameters.

Important for video segmentation and 3D scene understanding.

12. Bundle Adjustment

Refines 3D reconstruction by minimizing reprojection error.

Used in photogrammetry, aerial imaging, and Structure-from-Motion (SfM).

13. Watershed Segmentation

Treats image intensity as a topographic surface.

Watershed lines separate catchment basins (objects).

Effective for texture or region-based segmentation.

14. Road Marking Detection (ADAS / Vehicle Vision)

Techniques:

Edge detection + Hough Transform.

Color thresholding.

CNN-based segmentation (Deep Learning).

Used in lane keeping and autonomous driving.

15. Human Gait Analysis

Study of human walking pattern for biometric identification.

Features: joint angles, stride length, silhouette sequence.

Applications: surveillance, medical diagnosis, forensics.

16. Active Appearance Models (AAM)

Statistical model combining shape and texture variation.

Used in face recognition, tracking, and expression analysis.

Quick Recap Table