This repository contains my solutions for the two computer vision projects completed as part of the CS GY 6643 - Computer Vision course at NYU Tandon. Each project explored foundational techniques in computer vision, implemented from scratch without the use of pre-built libraries for the core algorithms.
-
Histogram Equalization
- Implemented functions for computing the Probability Density Function (PDF) and Cumulative Distribution Function (CDF) for image histograms.
- Enhanced image contrast by applying histogram equalization and analyzed its effects on image clarity and intensity.
-
Image Thresholding
- Applied Otsu's thresholding method to segment images and visualized inter-class variance.
- Explored the Niblack thresholding algorithm with different filter shapes to enhance cell boundaries in Transmission Electron Microscopy (TEM) images.
- Compared the effectiveness of Otsu’s and Niblack thresholding methods.
-
Template Matching
- Used cross-correlation to locate specific objects within a cluttered scene.
- Detected peaks in correlation maps and marked the identified locations on the original image.
- Analyzed results and discussed potential sources of false peaks.
-
Creative Section
- Developed an optimized approach for object detection to outperform standard cross-correlation in terms of execution time and efficiency.
-
Image Transformation and Stitching
- Aligned and stitched partial images to reconstruct a complete image resembling an A4 sheet.
- Compared manual feature selection and automated feature descriptors (e.g., SIFT, ORB) for image alignment and blending.
-
Hough Transform
- Detected straight lines to simulate runway detection for spacecraft landing.
- Implemented circular Hough Transform to identify circular landing pads, mimicking autonomous spacecraft navigation.
-
Image Segmentation
- Employed Mean Shift Segmentation and Normalized Graph Cut Segmentation to isolate stop signs in challenging scenarios.
- Discussed the strengths and weaknesses of both approaches in the context of road sign detection.
-
Creative Section
- Designed a custom segmentation algorithm leveraging classical techniques.
- Evaluated performance using Intersection over Union (IoU) as a metric and compared results with existing approaches.
Each project includes:
- Colab Notebooks: Implementation details and step-by-step analysis.
- Reports: Comprehensive explanations, results, and discussions.
Explore the repository for detailed implementations and insights into these computer vision techniques. Contributions and feedback are welcome!