Font size
WorksheetsContent Section Worksheet Questions (Grade 1 adapted MCQs)
Total questions: 100
Worksheet time: 50mins
A face detection system using Viola–Jones gives many false positives in cluttered backgrounds. Which change logically reduces this problem?
Increase image resolution
Add more stages to the cascade
Remove AdaBoost
Use grayscale images
YOLO detects objects faster than R-CNN. The main logical reason is that YOLO:
Uses fewer convolution layers
Avoids region proposals
Uses handcrafted features
Works only on videos
If YOLO fails to detect small objects, the most logical improvement is to:
Increase learning rate
Reduce number of classes
Increase grid resolution or use multi-scale detection
Convert images to grayscale
A CNN trained on daytime images performs poorly at night. The logical cause is:
Overfitting to illumination conditions
Poor optimizer
Incorrect labels
Insufficient epochs
Transfer learning improves performance when training data is limited because:
It avoids convolution
Pretrained models capture generic visual features
It reduces network depth
It increases batch size
Optical flow estimation becomes unreliable in textureless regions because:
Motion is zero
Brightness constancy fails
No intensity variation exists to track motion
Frames are identical
Horn–Schunck optical flow produces smoother motion fields than Lucas–Kanade because it:
Uses local windows
Enforces global smoothness constraint
Ignores brightness constancy
Uses feature matching
If fast-moving objects appear blurred, optical flow accuracy decreases mainly because:
Frame rate is low
Background is static
Motion is uniform
Image resolution is high
A GMM background model fails when lighting changes suddenly. The logical fix is to:
Increase number of Gaussians
Reduce learning rate adaptation time
Remove background subtraction
Convert to binary images
Multiple Gaussians per pixel are required because:
Foreground objects are static
Background may have repetitive variations
Noise is eliminated
Color images are used
If a pixel consistently appears in foreground for a long time, GMM will:
Always mark it as foreground
Ignore it permanently
Adapt and include it in background
Remove the pixel
Structure from Motion requires multiple images because depth cannot be inferred from:
Stereo images
Motion cues
A single static image
Feature correspondences
If camera motion is incorrectly estimated in SfM, the reconstructed 3D structure will be:
More accurate
Unaffected
Distorted
Faster to compute
Feature matching errors in SfM mostly affect:
Color consistency
3D point accuracy
Image resolution
Compression
Motion estimation is preferred over optical flow in video compression because it:
Is pixel-accurate
Is computationally simpler
Captures object semantics
Uses deep learning
Block-based motion estimation fails in regions with uniform intensity because:
No blocks exist
Matching cost becomes ambiguous
Motion is large
Noise is high
If Mean Squared Error is minimized in block matching, the algorithm favors:
Sharp edges
Maximum motion
Closest pixel similarity
Global motion
Image analysis focuses on feature extraction because raw pixels:
Are noisy
Are redundant and high-dimensional
Have low resolution
Cannot be processed
Texture features are better than color features when:
Objects have similar colors
Lighting is uniform
Images are grayscale
Resolution is high
Stereo vision depth estimation becomes inaccurate when baseline is too small because:
Disparity becomes negligible
Images overlap
Cameras misalign
Epipolar lines vanish
Large disparity corresponds to closer objects because:
Camera motion increases
Parallax effect is stronger
Images blur
Resolution increases
Epipolar constraint improves stereo matching efficiency by:
Reducing image size
Limiting search to a line
Increasing disparity
Removing noise
Motion models are required in animation to ensure:
Faster rendering
Realistic temporal consistency
Higher resolution
Color correction
Rigid motion models fail for human motion because:
Humans move slowly
Body parts deform independently
Images are noisy
Motion is linear
Affine motion models are preferred over simple translation because they can handle:
Noise
Scaling and rotation
Non-rigid deformation
Illumination change
A pattern recognition system fails mainly due to poor:
Display resolution
Feature selection
Camera quality
Storage capacity
Compact feature vectors are preferred because they:
Increase accuracy always
Reduce computation and overfitting
Improve brightness
Increase dimensions
Occlusion is a major challenge in object recognition because:
Objects disappear
Partial information is available
Lighting changes
Background is uniform
Normalization improves recognition accuracy by:
Increasing dataset size
Reducing irrelevant variations
Enhancing edges
Increasing contrast
Deep learning outperforms traditional methods mainly because it:
Uses rules
Learns hierarchical features
Requires no data
Avoids training
Overfitting occurs when a model performs well on training but poorly on testing due to:
Under-training
Excessive generalization
Memorization of training data
Poor labels
Data augmentation improves generalization by:
Reducing model size
Introducing controlled variability
Removing noise
Increasing epochs
Object recognition accuracy improves when motion cues are added because:
Color improves
Temporal consistency is exploited
Resolution increases
Noise is removed
A real-time system sacrifices some accuracy mainly to achieve:
Better training
Lower latency
Higher resolution
Larger models
Why is GMM preferred over simple thresholding in dynamic scenes?
Faster computation
Adaptive background modeling
Less memory
Simpler math
Stereo vision is superior to monocular vision for depth because it:
Uses learning
Uses geometry
Needs less data
Avoids calibration
If correspondence between stereo images is incorrect, depth estimation will be:
Accurate
Unchanged
Erroneous
Faster
Motion estimation improves tracking because it:
Reduces noise
Predicts object position
Enhances color
Increases resolution
In animation, non-rigid motion is necessary to model:
Vehicles
Buildings
Cloth and human skin
Cameras
Logical advantage of CNNs over traditional classifiers is their ability to:
Use fewer layers
Automatically learn spatial features
Avoid optimization
Eliminate training
If training and test data distributions differ, performance drops due to:
Overfitting
Dataset bias
Underflow
Noise removal
Combining stereo vision and motion estimation improves depth accuracy because:
Both reduce noise
Multiple cues reinforce estimation
Computation increases
Resolution improves
Feature redundancy negatively affects classifiers because it:
Improves accuracy
Increases computation without new information
Reduces data
Removes noise
Why are CNNs robust to small translations?
High learning rate
Pooling operations
Fully connected layers
Softmax
Logical drawback of very deep networks is:
Better accuracy
Vanishing gradients
Less data usage
Faster training
Motion blur affects object recognition because it:
Improves edges
Removes high-frequency details
Increases contrast
Stabilizes motion
Temporal consistency helps object recognition in video by:
Ignoring frames
Using previous frame information
Reducing frame rate
Removing motion
A model trained only on frontal faces fails on side views due to:
Low resolution
Viewpoint variation
Noise
Occlusion
Why is illumination invariance important in pattern recognition?
Lighting never changes
Objects appear different under different lighting
Images are grayscale
Cameras are fixed
Integrating detection, recognition, and tracking improves systems because it:
Increases complexity only
Uses complementary information
Reduces memory
Eliminates errors
In the Viola–Jones algorithm, integral images are primarily used to:
Reduce false positives using cascades
Compute Haar-like features in constant time
Improve image resolution
Classify images using CNNs
Why is AdaBoost critical in Viola–Jones face detection?
It enhances pixel contrast
It selects and weights the most discriminative features
It segments the image
It performs motion tracking
Which property of Haar-like features makes them suitable for real-time detection?
Rotation invariance
Low computational complexity
High texture sensitivity
Color dependence
The cascade classifier in Viola–Jones improves performance mainly by:
Using deep neural networks
Rejecting negative regions early
Increasing training data
Applying frequency-domain filtering
YOLO treats object detection as a:
Classification problem
Region proposal problem
Regression problem
Clustering problem
Why does YOLO achieve faster detection compared to R-CNN-based methods?
Uses handcrafted features
Processes the entire image in a single forward pass
Uses smaller datasets
Avoids convolution operations
In YOLO, grid cells are responsible for:
Image enhancement
Motion estimation
Predicting bounding boxes and class probabilities
Stereo correspondence
A limitation of YOLO is its difficulty in detecting:
Large objects
Textured objects
Small objects close together
Static objects
Which deep learning architecture is most commonly used for object recognition?
RNN
CNN
GAN
Autoencoder
CNNs are effective for object recognition mainly because they:
Use handcrafted features
Exploit spatial hierarchy using convolution
Reduce image size only
Perform unsupervised clustering
Which layer in a CNN provides translation invariance?
Convolution layer
Fully connected layer
Pooling layer
Softmax layer
Transfer learning is useful in object recognition because it:
Eliminates training
Reuses learned features from large datasets
Reduces image noise
Improves optical flow accuracy
Optical flow estimates:
Object boundaries
Pixel-wise motion between frames
Image depth
Color consistency
The brightness constancy assumption in optical flow states that:
Brightness changes over time
Pixel intensity remains constant along motion
Motion is linear
Objects do not move
Gaussian Mixture Models (GMM) are commonly used in vision for:
Edge detection
Background subtraction
Stereo matching
Object classification
In GMM-based background modeling, each pixel is represented as:
Single Gaussian
Binary value
Mixture of Gaussians
Histogram
Why are multiple Gaussians used per pixel in GMM?
To reduce noise
To model complex background variations
To enhance contrast
To increase frame rate
Foreground objects in GMM are detected when pixel values:
Match dominant Gaussians
Do not fit background distributions
Have high brightness
Are stationary
Structure from Motion (SfM) aims to recover:
Object color
Camera calibration only
3D structure and camera motion
Optical flow vectors
SfM primarily relies on:
Single image analysis
Multiple views of a scene
Stereo pairs only
Infrared images
Which feature is commonly used in SfM pipelines?
Haar features
SIFT / ORB
Histogram equalization
Thresholding
Motion estimation differs from optical flow because it:
Is pixel-based
Estimates block or global motion
Uses color models
Detects objects
Block-based motion estimation is widely used in:
Object recognition
Image enhancement
Video compression
Stereo vision
Which metric is commonly minimized in block matching algorithms?
Entropy
Mean Squared Error
Precision
Recall
Image analysis in pattern recognition primarily focuses on:
Image compression
Extracting meaningful features
Reducing resolution
Color correction
Which feature is most suitable for texture analysis?
Color histogram
Edge orientation
GLCM features
Bounding box size
Stereo vision estimates depth using:
Motion vectors
Disparity between image pairs
Optical flow
GMM
Greater disparity in stereo vision indicates:
Object is farther
Object is closer
No depth
Flat surface
The epipolar constraint reduces correspondence search to:
Entire image
Horizontal line
Vertical line
Epipolar line
Which factor most affects stereo matching accuracy?
Image brightness
Camera baseline
Color depth
Compression ratio
Motion models for image animation describe:
Pixel colors
Object movement over time
Image resolution changes
Noise patterns
Rigid motion assumes that:
Object shape changes
Object deforms
Distances between points remain constant
Motion is random
Which is an example of non-rigid motion?
Rotating wheel
Translating car
Walking human
Moving box
Affine motion model can represent:
Only translation
Translation and rotation only
Translation, rotation, scaling, shear
Non-linear deformation
Why are motion models important in animation?
Reduce storage
Produce realistic movement
Enhance brightness
Remove noise
Pattern recognition systems typically consist of:
Sensor, feature extractor, classifier
Encoder, decoder, filter
Camera, display, memory
Histogram, threshold, output
Feature vectors should ideally be:
High dimensional
Redundant
Discriminative and compact
Random
Which classifier is commonly used in object recognition tasks?
KNN
SVM
Decision Tree
All of the above
Why is normalization important in pattern recognition?
Increase image size
Reduce variation due to scale and illumination
Improve color
Detect motion
Which challenge is common in real-world object recognition?
Uniform background
Controlled lighting
Occlusion
Fixed viewpoints
Deep learning reduces the need for manual feature design because it:
Uses predefined rules
Learns features automatically
Uses templates
Ignores data
Overfitting in deep object recognition models occurs when:
Model generalizes well
Training loss is high
Model memorizes training data
Dataset is large
In the context of motion estimation, which technique is primarily used to track moving objects across frames?
Image segmentation
Optical flow
Color histogram
Feature extraction
What is the primary advantage of using convolutional layers in a CNN for image processing?
They increase image resolution
They simplify the model architecture
They reduce the number of parameters
They enhance color accuracy
In stereo vision, the disparity map is crucial for:
Identifying object boundaries
Estimating depth information
Enhancing image contrast
Reducing noise in images
What is the primary advantage of using convolutional neural networks (CNNs) for image classification?
They require less data preprocessing
They are easier to implement
They are faster than traditional methods
They automatically learn spatial hierarchies
In the context of image segmentation, what does the term 'superpixel' refer to?
A pixel that has been compressed
A group of pixels with similar colors
A pixel with enhanced brightness
A pixel that is part of a larger object
Which technique is commonly used to improve the robustness of feature matching in varying lighting conditions?
Histogram equalization
Edge detection
Image blurring
Color inversion
What is the primary advantage of using Convolutional Neural Networks (CNNs) for image classification?
They require less data preprocessing
They can learn spatial hierarchies of features
They eliminate the need for feature extraction
They are faster than traditional algorithms
In the context of image segmentation, what does the term 'pixel-wise classification' refer to?
Classifying entire images
Reducing image resolution
Assigning a class label to each pixel
Detecting edges in images
