Image Segmentation Project

Overview

This project is part of design and analysis of algorithms course that implements an image segmentation algorithm using a graph-based approach inspired by the Felzenszwalb-Huttenlocher method. The application is built in C# using Windows Forms, allowing users to load an image, apply Gaussian smoothing, segment the image into regions based on color similarity, and visualize the results. The segmentation process leverages a Disjoint Set Union (DSU) data structure for efficient component merging and supports parallel processing for red, green, and blue channels.

Features

• Image Loading and Display: Load images in various formats (24-bit RGB, 32-bit ARGB, 8-bit indexed) and display them on a PictureBox.
• Gaussian Smoothing: Apply a 1D Gaussian filter to enhance edge detection, with configurable sigma and filter size.
• Graph-Based Segmentation: Segment images by constructing a graph where pixels are nodes and edges represent color differences, using a DSU for efficient merging.
• Parallel Processing: Process red, green, and blue channels concurrently to improve performance.
• Output Visualization: Display the segmented image with random colors assigned to each segment and save segment sizes to a text file.
• User Interface: Windows Forms interface with controls for specifying sigma and k parameters, loading images, and viewing results.

Algorithm Description

The segmentation algorithm follows these steps:

1. Image Loading: The input image is loaded into a 2D array of RGBPixel structs.
2. Gaussian Smoothing: A 1D Gaussian filter is applied to the image to reduce noise and enhance edge detection, controlled by the sigma parameter.
3. Graph Construction: A graph is built where each pixel is a node, and edges connect neighboring pixels (down, down-left, down-right, right) with weights based on color differences.
4. Channel-Wise Segmentation: Each color channel (red, green, blue) is segmented independently using the DSU:
   • Edges are sorted by weight using counting sort (weights range from 0 to 255).
   • For each edge, components are merged if the edge weight is below a dynamic threshold, controlled by the k parameter.
5. Final Segmentation: Pixels with identical label triplets across all channels are merged into final segments using another DSU pass.
6. Output Generation: Segments are assigned random colors, and the segmented image is saved as a BMP file. Segment sizes are written to a text file and displayed in the UI.

Usage

1. Run the Application: Launch the program to open the Windows Forms interface.
2. Load an Image: Click the "Open" button to select an image file.
3. Set Parameters:
   • Sigma: Controls the Gaussian filter's spread (default: 0.8).
   • k: Controls the segmentation granularity (default: 300; higher values produce larger segments).
4. Segment the Image: Click the "Segment" button to process the image.
5. View Results: Click the "Show" button to display the segmented image and segment sizes. The segmented image is saved as tempImage.bmp, and segment sizes are saved as tempText.txt.

Dependencies

• .NET Framework (Windows Forms for UI)
• System.Drawing for image processing
• No external libraries required

Performance Considerations

• Parallel Processing: The use of Parallel.Invoke speeds up channel-wise segmentation.
• Counting Sort: Optimizes edge sorting for weights in the range [0, 255].
• Path Compression in DSU: Improves the efficiency of find operations in the Disjoint Set Union structure.

Building and Running

1. Clone the repository:

   git clone "https://github.com/David-Magdy/Image-Segmentation"

2. Open the solution in Visual Studio.
3. Build and run the project to launch the Windows Forms application.

Example Output

• Segmented Image: A BMP file (tempImage.bmp) with each segment colored randomly.
• Segment Sizes: A text file (tempText.txt) listing the number of segments followed by their sizes in descending order.
• UI Display: The segmented image is shown in a PictureBox, and segment sizes are displayed in a multiline TextBox.

Screenshots

Below are example images showing the input and output of the segmentation process:

• Example 1

  

• Example 2

  

• Example 3

  

License

This project is licensed under the MIT License.