cv_fundamentals

This repository aims to be an educative resource covering interactive implementations of the founding concepts of Computer Vision taught in CS5613: Computer Vision Fundamentals at LUMS, Pakistan.

Installation

Clone the repository and install the required packages with:

pip3 install -r requirements.txt

Filter Banks

Click to expand

In Computer Vision, filter banks involve sets of filters used for image processing tasks like edge detection, blurring, or sharpening, often applied through convolution operations. These banks comprise various filters that isolate specific features within an image, employing convolution to modify or analyze the image for tasks such as enhancement or feature extraction.

The code in the folder filter bank is an Image Filter App built using Streamlit and OpenCV. It allows users to upload an image and allows to play around and experiment with applying and combining various filters. Listed below are the available filters:

Filter Name	Description
Grayscale	Converts the image to grayscale, removing color information.
Blur	Applies Gaussian blur, smoothing out details and reducing noise.
Edge Detection	Detects edges using Canny algorithm, highlighting object boundaries.
Sepia	Imparts a warm, brownish tone, resembling old-fashioned photos.
Invert Colors	Inverts the colors of the image, creating a negative effect.
Sharpen	Enhances the contrast along edges, making the image appear sharper.
Emboss	Creates a raised or engraved appearance by emphasizing edge differences.
Cartoonize	Transforms the image into a cartoon-like representation.
Box Blur	Applies a box blur effect, averaging pixel values in a defined area.

1. Run Streamlit App using the command streamlit run _main.py and upload an image.

2. Apply a filter: Check a filter from the list to apply to the image.

3. Apply Multiple Filters: Check multiple boxes to create a compounding effect.

4. TO-DO: Increase diversity in the filter bank.

Vanishing Point

Click to expand

Vanishing points refer to the points in a perspective image where parallel lines appear to converge or intersect. They are the foundation of projective geometry and play a major role in understanding the spatial relationships between objects in a scene, aiding in tasks like object detection, 3D reconstruction, and image rectification. In Stanley Kubrick's films like "The Shining" and "2001: A Space Odyssey," his use of one-point perspective illustrates this concept vividly.

The code provided in the folder vanishing point utilizes Tkinter for the GUI, and OpenCV for image handling. Here's a brief overview of how to interact with it:

1. Run "_main.py" to open a window with an "upload file" button. Upload an image to display it.:

2. Mark 4 Points: Another window will open after image upload. Mark points for two converging parallel lines in order.

3. Vanishing Point: The updated image will show the vanishing point marked in green.

4. TO-DO: Implement vanishing points calculation for 2-point and 3-point perspective.

Image Stitching

Click to expand

In image stitching, homography serves as the cornerstone. It's a transformation matrix that maps points from one image to another, enabling alignment and blending. By identifying corresponding points in multiple images, the homography matrix helps warp and merge these images seamlessly. This process aligns the images according to the perspective from which they were taken, ultimately creating a panoramic or stitched image. Atleast 4 points are required to find the transformation matrix.

Note: The following implementation is purely based on the equations above ⬆️. Correspondence error correction techniques like RANSAC have not been incorporated so kindly mark your correspondences with precision.

The code in the folder stitch_images is an Image Stitching App built using Streamlit and OpenCV. It enables users to upload two images and select corresponding points on these images. The app applies homography to stitch the images together based on the chosen points to create a panorama-like image.

1. Run Streamlit App using the command streamlit run _main.py

2. Image Upload Upload two images.

3. Point Selection: Mark atleast 4 corresponding points on both images alternatively.

4. Stitching: The app calculates a homography matrix based on the selected points and stitches the images together to create a panoramic view.

Structure From Motion (SFM)

Click to expand