MCS-230: Digital Image Processing and Computer Vision
Q1
Image acquisition is the process of capturing an image from the real-world environment using imaging
devices. It is the first step in image processing.
- Optical Image Processing: Uses light and optical components to manipulate images directly.
- Analog Image Processing: Processes continuous signals using electronic circuits; susceptible to noise.
- Digital Image Processing: Converts images to digital form and uses algorithms for enhancement,
restoration, etc. It offers better flexibility, reproducibility, and storage.
Q2
First, calculate physical size in mm: 4 inches = 101.6 mm.
- Number of cycles = 5 cycles/mm * 101.6 mm = 508 cycles.
- Pixels required = 2 * 508 = 1016 pixels per side.
- Conclusion: An image of 512x512 is not sufficient to preserve all details.
Q3
Types of Images:
- Based on Attributes:
- Binary Images: Two pixel values (0 or 1).
- Grayscale Images: Shades of gray.
- Color Images: Multiple channels like RGB.
- Based on Colour:
- Monochrome: Black and white.
- RGB, CMY, HSV: Different models representing color information.
Q4
(a) Storage for 2024x2024 24-bit image: 12.29 MB.
(b) Pixel resolution for 3000x4000 image: 12 Megapixels.
(c) Grayscale image with aspect 8:2 and 1 million pixels: Dimensions = 2000x500, Size ~ 1 MB.
Q5
Enhancement in Frequency Domain:
MCS-230: Digital Image Processing and Computer Vision
- Selective enhancement/suppression of features.
- Noise reduction becomes simpler.
- Global image properties are better handled compared to the Spatial Domain.
- Fourier Transform allows efficient implementation.
Q6
Smoothing Filters:
- ILPF: Passes all frequencies below cutoff sharply; causes ringing.
- BLPF: Smooth transition around cutoff; avoids sharp artifacts.
- GLPF: Smoothest filter; no ringing, best quality smoothing.
Q7
Sharpening Filters:
- IHPF: Blocks low frequencies; allows high frequencies; causes ringing.
- BHPF: Gradual transition; controlled sharpening.
- GHPF: Smooth sharpening without ringing artifacts.
Q8
- Mean Filter: Averages surrounding pixels; good for Gaussian noise.
- Median Filter: Takes the median value; excellent for salt-and-pepper noise removal.
Example: In a 3x3 matrix, replace center pixel by average (mean) or middle (median) value.
Q9
RGB to CMY Transformation:
- Each CMY component = 1 - corresponding RGB component.
- Vertices labeled accordingly: Black, Cyan, Magenta, Yellow, etc.
- High saturation colors lie at the corners; low saturation towards the center.
Q10
Optical Flow:
- Pattern of motion between frames.
MCS-230: Digital Image Processing and Computer Vision
- Each pixel's motion vector calculated.
- Applications: Motion detection, object tracking, 3D structure estimation.
- Lucas-Kanade and Horn-Schunck are popular methods.
Q11
Epipolar Geometry:
- Defines relation between two camera views.
- Key concepts: Epipoles, Epipolar Lines, Baseline.
- Reduces search for matching points to 1D search along epipolar lines.
- Critical for stereo vision and depth estimation.
Q12
Camera Calibration:
- Determines intrinsic (focal length, optical center) and extrinsic (rotation, translation) parameters.
- Uses known patterns (checkerboards) for estimation.
- Necessary for accurate 3D reconstruction and rectification.
Q13
K-means Clustering:
- Assigns data into k clusters by minimizing variance within clusters.
- Advantages: Simple, fast.
- Disadvantages: Requires initial k, sensitive to initial placement.
- Example: Grouping points based on distance to centroids.
Q14
Frogy's Method (k=2, seed points (3,3) and (6,8)):
- Assign points to nearest centroid based on distance.
- Update centroids iteratively.
- Final clusters formed based on updated centroids after stabilization.
Q15
MCS-230: Digital Image Processing and Computer Vision
Agglomerative Hierarchical Clustering:
- Bottom-up approach; each point starts as its own cluster.
- Types of Linkage:
- Single Linkage: Nearest points.
- Complete Linkage: Farthest points.
- Average Linkage: Mean distance.
- Dendrogram is used for visualization.
Q16
Bayes Classifier:
- Uses Bayes' Theorem to assign probabilities to classes.
- Chooses class with maximum posterior probability.
- Properties: Assumes independence, simple, fast, good for text classification and spam detection.
