Font size
WorksheetsITA0511 – Computer Vision for Image Processing: Class Test 3
Total questions: 100
Worksheet time: 50mins
A 512×512 image is downsampled to 256×256 without anti-aliasing. The most likely visual artifact is:
Blurring
Aliasing
Quantization noise
Contrast loss
If spatial resolution is doubled in both x and y directions, total number of pixels increases by:
2×
3×
4×
8×
Reducing intensity levels from 256 to 16 primarily increases:
Spatial noise
Quantization error
Blur
Edge thickness
A 10-bit grayscale image has how many intensity levels?
256
512
1024
2048
Which operation improves intensity resolution but not spatial resolution?
Sampling
Quantization
Interpolation
Smoothing
For highlighting low-intensity details in an image, the best transformation is:
Histogram equalization
Log transformation
High-pass filtering
Negative transformation
Histogram equalization may fail when:
Image has uniform histogram
Image has local contrast variations
Image is grayscale
Image is large
Adaptive histogram equalization is preferred because it:
Reduces noise
Works on local regions
Is faster
Preserves edges
Increasing y (>1) in power-law transformation causes the image to:
Brighten
Darken
Blur
Sharpen
Contrast stretching mainly modifies:
Frequency components
Intensity range
Spatial resolution
Image size
A median filter is best for removing:
Gaussian noise
Salt-and-pepper noise
Periodic noise
High-frequency noise
Gaussian smoothing is applied before edge detection to:
Strengthen edges
Reduce noise sensitivity
Increase resolution
Improve contrast
Mean filter causes edge blurring because it:
Is non-linear
Averages neighboring pixels
Amplifies gradients
Detects corners
Laplacian filter enhances edges but is sensitive to noise because it is:
Low-pass
Non-linear
Second-order derivative
Binary
Which filter preserves edges while smoothing flat regions?
Mean
Median
Gaussian
Laplacian
Low-pass filtering in frequency domain removes:
Edges
Noise
Smooth regions
Texture
DCT is preferred in JPEG because it provides:
Lossless compression
Energy compaction
Edge detection
Noise suppression
Removing all high-frequency components results in:
Sharpened image
Blurred image
No change
Binary image
Most image energy is concentrated in:
High frequencies
Mid frequencies
Low frequencies
Random frequencies
Frequency domain enhancement operates on:
Pixels
Neighborhoods
Transform coefficients
Gradients
Edge detection is based on detecting:
Intensity values
Zero crossings
Rapid intensity changes
Color variation
Sobel operator gives stronger edges than Prewitt because it:
Uses larger kernel
Includes smoothing
Uses frequency domain
Is non-linear
Canny detector uses double thresholding to:
Reduce noise
Remove false edges
Laplacian edge detector often produces double edges because it:
Uses gradients
Detects zero crossings
Is non-linear
Uses thresholding
Which edge detector is most robust to noise?
Roberts
Prewitt
Sobel
Canny
Global thresholding fails when the image has:
Uniform background
High contrast
Varying illumination
Binary regions
Region-based segmentation groups pixels based on:
Edge strength
Similarity
Gradient direction
Noise level
Watershed segmentation often causes over-segmentation due to:
Noise and local minima
Poor edges
Low resolution
Thresholding
Edge-based segmentation may fail when:
Objects have smooth boundaries
Edges are weak or broken
Image is binary
Contrast is high
Segmentation is application-dependent because:
Images vary
Objects vary in appearance
No universal method exists
All of the above
Corners are preferred features because they have:
High intensity
High variation in all directions
Uniform gradient
Smooth texture
Harris detector fails on edges because:
Gradients exist in one direction
Noise dominates
Eigenvalues are equal
Threshold is high
SIFT features are invariant to scale due to:
Normalization
Gaussian pyramid
Histogram equalization
Thresholding
PCA reduces dimensionality by selecting components with:
Minimum variance
Maximum variance
Minimum mean
Maximum entropy
Retaining fewer PCA components mainly removes:
Important features
Redundant information
Edges
Noise only
Morphological erosion primarily:
Expands objects
Shrinks objects
Smooths edges
Detects corners
Opening operation is useful for removing:
Large objects
Thin noise
Edges
Texture
Closing operation is used to:
Remove small holes
Detect boundaries
Sharpen edges
Reduce noise
CNNs outperform classical methods because they:
Use fixed filters
Learn hierarchical features
Reduce noise
Use PCA
Pooling layers in CNNs mainly reduce:
Depth
Feature map size
Kernel size
Noise
Image preprocessing improves:
Storage only
Noise only
Feature extraction accuracy
Image size
Feature extraction follows segmentation because:
Regions are defined
Noise is removed
Image is resized
Histogram is flattened
Object recognition accuracy depends heavily on:
Camera resolution
Feature quality
File format
Image size
Real-time vision systems prefer SURF over SIFT due to:
Accuracy
Robustness
Speed
Memory
PCA before classification mainly reduces:
Training time
Image contrast
Noise only
Pixel count
A 256×256 image divided into 8×8 blocks produces:
512 blocks
1024 blocks
2048 blocks
4096 blocks
Quantizing 256 levels into 16 levels gives a step size of:
8
16
32
64
If PCA retains 95% variance with 40 components, reducing to 20 components will most likely:
Increase accuracy
Slightly reduce accuracy
Improve contrast
Increase noise
Removing low-frequency components affects the image by:
Removing edges
Removing smooth regions
Increasing brightness
Reducing noise
The first step in a computer vision pipeline is:
Segmentation
Feature extraction
Image acquisition
Classification
A digital image is represented as:
Continuous signal
Discrete intensity values
Binary waveform
Analog voltage
Image preprocessing mainly aims to:
Reduce file size
Improve image quality
Classify objects
Detect feature
Noise in images generally affects:
Low frequencies
High frequencies
Color depth
Image resolution
Salt-and-pepper noise is best removed using:
Mean filter
Median filter
Gaussian filter
Laplacian filter
Gaussian smoothing is mainly used to:
Enhance edges
Remove noise
Detect corners
Segment images
Histogram equalization improves:
Brightness
Contrast
Resolution
Noise
Contrast stretching is a:
Frequency domain technique
Point processing technique
Segmentation method
Compression technique
Negative image transformation is useful for:
Medical images
Satellite images
Low contrast images
All of the above
Adaptive histogram equalization works on:
Entire image
Fixed window size
Local regions
Edge pixels
Which transformation highlights dark regions?
Log transformation
Power-law (γ < 1)
Histogram equalization
High-pass filter
Mean filter is mainly used for:
Edge detection
Noise smoothing
Image sharpening
Feature extraction
High-pass filtering emphasizes:
Smooth regions
Low frequencies
Edges and details
Background
Laplacian filter is a:
First-order derivative
Second-order derivative
Zero-order filter
Non-linear filter
Which filter preserves edges while smoothing?
Mean filter
Median filter
Gaussian filter
Box filter
Sobel operator computes:
Laplacian
Gradient magnitude
Histogram
Frequency response
Fourier Transform converts image from:
Spatial to frequency domain
Frequency to spatial domain
Pixel to binary
RGB to grayscale
Low-pass filtering removes:
Smooth regions
Noise
Edges
Texture
High-pass filters are used for:
Blurring
Smoothing
Sharpening
Compression
DCT is widely used in:
PNG
JPEG
BMP
TIFF
Frequency domain enhancement operates on:
Pixels
Gradients
Coefficients
Neighborhoods
Edge detection finds:
Object interiors
Object boundaries
Color variation
Texture patterns
Canny edge detector is preferred because it:
Is fast
Reduces noise and detects edges accurately
Uses simple masks
Needs no thresholds
Thresholding is a:
Enhancement method
Segmentation technique
Compression technique
Feature descriptor
Global thresholding fails when:
Image is noisy
Illumination is uniform
Region-based segmentation groups pixels based on:
Edges
Similarity
Gradients
Noise
Harris corner detector is based on:
Gradient magnitude
Eigenvalues of matrix
Histogram
Frequency response
SIFT features are invariant to:
Rotation and scale
Noise only
Translation only
Illumination only
SURF improves over SIFT by:
Accuracy
Speed
Robustness
Compression
Corner points have:
High intensity
High variation in all directions
Uniform gradient
Zero gradient
Feature descriptors are used for:
Noise removal
Image matching
Image compression
Image acquisition
PCA reduces:
Image noise
Data dimensionality
Image resolution
Image brightness
Principal components are selected based on:
Mean
Variance
Correlation
Entropy
Higher eigenvalues represent:
Noise
Less information
More information
Background
DCT concentrates energy in:
High-frequency coefficients
Low-frequency coefficients
Random coefficients
Edge coefficients
Compression ratio is defined as:
Output / Input
Input / Output
Bits per pixel
Pixel count
Medical image enhancement is critical for:
Entertainment
Diagnosis
Compression
Transmission
Computer vision differs from image processing because it focuses on:
Pixels
Visual understanding
Filtering
Enhancement
Image segmentation is performed before:
Image acquisition
Feature extraction
Image sensing
Sampling
Sampling affects:
Image resolution
Image contrast
Image noise
Image color
Quantization affects:
Spatial resolution
Intensity resolution
Image size
Image orientation
Over-segmentation is common in:
Edge-based methods
Region growing
Watershed algorithm
Thresholding
Morphological operations are based on:
Frequency analysis
Set theory
Linear algebra
Probability
Dilation operation:
Shrinks objects
Expands objects
Detects edges
Removes noise
Erosion operation:
Thickens objects
Smooths edges
Shrinks objects
Detects corners
Opening operation is erosion followed by:
Thresholding
Dilation
Closing
Filtering
First stage of vision pipeline:
Recognition
Feature extraction
Image acquisition
Classification
Object recognition requires:
Segmentation only
Feature extraction and classification
Filtering only
Thresholding only
CNNs are effective because they:
Use handcrafted features
Learn hierarchical features
Use PCA
Use DCT
Pooling layers reduce:
Noise
Feature map size
Color depth
Kernel size
Computer vision is widely used in:
Autonomous vehicles
Medical imaging
Surveillance
All of the above
