Photo Mosaic

Concept

A photo mosaic algorithm built to compose a large base image entirely out of a library of smaller source photographs. The tool was developed to generate massively detailed outputs, enabling very high-resolution final renders suitable for large-scale printing.

Technical Implementation

The system is written in C++ using the openFrameworks creative coding toolkit and OpenCV (via the ofxCv addon) for robust image processing.

• Color Analysis & Grid Processing: The algorithm begins by dividing the target (base) image into a grid of uniform tiles. For each tile, it computes the mean RGB color value using OpenCV’s cv::mean function. It performs the exact same analysis on an entire directory of source images, storing their mean colors in memory.
• Distance-Based Matching: To find the best fitting photograph for a specific tile, the system calculates the Euclidean distance (cv::norm) between the color vectors of the target tile and each source image.
• Repetition & Diversity Control: A naive matching approach often results in the same few images being repeated in areas of uniform color. To solve this, the algorithm employs a clever two-pass priority queue mechanism. First, it identifies the top 20 closest matches by color distance. From those 20 candidates, it selects the one with the lowest usedCount (the image that has been placed the least number of times). This forces a visually diverse distribution of photos across the final mosaic.
• High-Resolution Tiling Engine: Generating a massive mosaic in memory can easily cause out-of-memory errors. The tool mitigates this by splitting the final output into smaller, manageable sub-regions (e.g., a 9x9 grid). Each section is processed, populated with resized source images, and individually saved to disk, allowing for virtually unlimited final output resolutions.