Color-Based Image Similarity

A Windows Forms application that finds similar images based on color histogram analysis. This project was developed as part of an algorithms course contest, where it achieved 9th place out of 50 participants.

Overview

This application uses color histogram analysis to measure image similarity. It calculates statistical features from RGB color channels and uses cosine distance to find the most similar images in a dataset.

Features

• Color Histogram Analysis: Calculates comprehensive statistics for each RGB channel:

  • Histogram (256 bins for each channel)
  • Minimum, Maximum, Median values
  • Mean and Standard Deviation

• Image Similarity Matching: Uses cosine distance between probability distributions of color histograms to find similar images

• Parallel Processing: Leverages multi-threading to efficiently process multiple images simultaneously

• Interactive GUI:

  • Load a folder of target images
  • Select a query image
  • Find top N most similar images
  • Visualize RGB histograms for both query and matched images

• Performance Metrics: Displays loading and matching times

Algorithm

The similarity matching algorithm works as follows:

1. Image Statistics Calculation: For each image, compute:

   • RGB histograms (256 bins per channel)
   • Statistical measures: min, max, median, mean, standard deviation

2. Similarity Scoring: For each target image, calculate cosine distance:

   • Convert histograms to probability distributions
   • Compute cosine distance for each RGB channel separately
   • Average the three channel distances to get the final similarity score

3. Top Matches: Sort all images by similarity score and return the top N matches

The cosine distance formula used:

distance = arccos(dot_product / (norm1 * norm2)) * (180 / π)

Where the final score is the average of R, G, and B channel distances.

Requirements

• .NET Framework 4.8
• Windows OS
• Visual Studio (for building from source)

Building the Project

1. Open ImageSimilarity.sln in Visual Studio
2. Build the solution (F6 or Build → Build Solution)
3. Run the application (F5)

Usage

1. Load Images: Click "Load Images" and select a folder containing images

   • Supported formats: JPG, JPEG, PNG, BMP, GIF, TIFF, ICO
   • The application will calculate statistics for all images in parallel

2. Select Query Image: Click "Open" to select the image you want to find matches for

   • The query image and its RGB histograms will be displayed

3. Find Matches:

   • Set the number of top matches desired (using the numeric up/down control)
   • Click "Match Image" to find the most similar images
   • Results will be displayed in the matched images list

4. View Results:

   • Click on any matched image to view it and its histograms
   • The match score (lower = more similar) is displayed

Project Structure

• ImageHistSimilarity.cs: Core algorithm for calculating image statistics and finding matches
• ImageOperations.cs: Image I/O operations and histogram visualization
• MainForm.cs: GUI implementation and user interaction handling
• Program.cs: Application entry point

Performance Optimizations

• Parallel Image Loading: Uses Parallel.For to process multiple images concurrently
• Unsafe Code: Direct memory access for efficient image pixel reading
• Single-Pass Statistics: Calculates all statistics in a single iteration through the image

Contest Achievement

This project was submitted as part of an algorithms course contest and achieved 9th place out of 50 participants, demonstrating effective implementation of color-based image similarity algorithms.

License

This project was created for educational purposes as part of an algorithms course.