Description
1 Image Transformation and Stitching
After the midterms, the teaching assistants (TAs) were occupied with scanning the submitted answer
sheets. In their rush to attend their next class, they mistakenly captured only partial images of each
sheet. Below are the three photos they took in haste.
Help the TAs rectify this mistake by combining these images to produce a complete, rectangular image
that displays all parts of the answer sheet, resembling an A4 sheet as closely as possible.
(a) Image 1 (b) Image 2 (c) Image 3
Figure 1: Scanned image sections
Suggested Approach
• Feature Identification: Determine the location of corresponding features between the images,
which will facilitate accurate alignment and stitching. This can be done in one of the following
ways:
– Manual Identification: Select corresponding points between the images by hand. (3)
– Automated Detection: Use a feature descriptor, such as SIFT or ORB, to automatically
identify matching points across the images. (5)
• Image Stitching: Using the identified points, align and stitch the images to form a composite.
(6)
• Rectification: Correct any perspective distortions and non-uniform scaling to ensure that the
final image is rectangular, resembling an A4 sheet. (6)
• Comparative Analysis: Evaluate and discuss the quality of stitching obtained through manual feature selection versus using an automated feature descriptor. Comment on the accuracy,
smoothness of blending, and ease of implementation for each method.(5)
2 Hough Transform
After the space shuttle has completed its mission in orbit, it begins its descent back to Earth. As it
approaches the atmosphere, its navigation systems activate advanced visual processing techniques to
prepare for a precise landing. Upon entering the atmosphere, the shuttle’s sensors scan for a safe landing
runway. Using Hough Transform algorithms, the shuttle identifies straight lines on the runway, which
helps it align perfectly for touchdown.
Figure 2
Your mission is to simulate the space shuttle’s final approach. Given an image representing the shuttle’s
view of the landing area, use Hough Transform to detect and delineate the runway lines. This will aid
in guiding the shuttle for a safe landing.
Figure 3
In a groundbreaking mission, SpaceX’s Falcon 9 booster launches a payload into orbit. After the booster
separates, it autonomously begins its descent, aiming to land precisely on a circular landing pad on land.
During its descent, the booster’s cameras and sensors analyze the view of the landing zone searching for
the landing pad. To ensure a safe landing, the booster scans the surface, using a powerful algorithm to
detect the circular landing zone amidst other infrastructure and clutter.
Your mission is to assist the Falcon 9 booster’s landing process. Provided with an image 5 that mimics
the booster’s view as it nears the landing zone, use Hough Transform to locate the circular landing pad.
Accurate detection of this circle is crucial for a safe touchdown.
Figure 4
Figure 5
Suggested Approach
• Implementation: Implement the Hough transform from scratch for both the scenarios
– Edges of runway: Identify the bounding edges of the runway for precise touchdown and
landing (7)
– Edges of landing pad: Identify the both the bounding edges of the circular(the smaller
circle and the larger circle) Landing pad for precise touchdown and landing (7)
• Overlaying: Using the implementation of hough transform function overlay the final result on the
two original images (3+3)
• Improvements: Discuss if the detection is satisfactory if not what changes would you suggest to
make the mission successful (5)
3 Segmentation
After the successful completion of various space missions, you are now traveling in your car equipped
with Full Self-Driving (FSD) capabilities. However, despite its sophisticated algorithms for autonomous
driving, the system is glitching when it comes to identifying stop signs—a critical aspect of safe navigation
on the road. As a computer vision expert, you are tasked with implementing a quick and effective solution
to resolve this issue using image segmentation techniques.
Objectives
Your goal is to perform image segmentation on a given image 6 that include a stop sign. You will employ
two different methods of image segmentation and compare their effectiveness.
• Method 1: Mean Shift Segmentation
– Implement the Mean Shift algorithm to segment the images. This technique groups pixels
based on color and spatial proximity, which can help in isolating the stop signs from the
background.
• Method 2: Normalized Graph Cut Segmentation
– Utilize the Normalized Graph Cut method for segmentation. This method formulates the
image as a graph and partitions it based on the minimum cut, allowing for better delineation
of complex structures within the image, including stop signs.
Figure 6
Deliverables
• For each segmentation method, provide:
– A brief description of the algorithm and its implementation. (10)
– The original image and the segmented result for comparison. (10)
– A discussion on the strengths and weaknesses of each method in the context of identifying
stop signs. (5)
4 Creative Section
Points (25)
If you beat this section, you will get (TBD) bonus points.
The task of this creative section is to come up with the best segmentation algorithm which is not
machine/deep learning based approach and follows classical approaches.
4.1 Task Description
• Image Selection: Choose a challenging image for segmentation. The complexity of the image
will provide a better opportunity to showcase your algorithm’s effectiveness.
• Performance Evaluation: – Calculate the Intersection over Union (IoU) values for all segmented
objects within the image. This metric will provide a quantitative measure of segmentation accuracy,
defined as:
IoU = Area of Overlap
Area of Union
– Report the IoU values for each identified segmented region.
• Detailed Approach Description: Provide a comprehensive explanation of the approach you
implemented, including: – The rationale behind your choice of algorithm. – Step-by-step details of
the segmentation process. – A justification of why you believe this method performs better than
alternative approaches.
• Comparison and Empirical Evidence: Compare your results with other classical segmentation
techniques (if applicable) and present empirical proof of your method’s performance. This may
include visual comparisons, numerical results, and any relevant metrics that support your claims.
• Function Requirements: Your segmentation algorithm should be encapsulated in a function
that accepts an image as input and outputs all identified segmented regions.
• Bonus Eligibility: The top TBD% of students who achieve the highest average IoU for all
identified regions will be eligible to receive the bonus points.
