Pixi-Collage

Pixi-Collage is a Python tool that generates artistic image collages by recreating an input image using frames extracted from a video. Each pixel in the original image is replaced with a video frame that has a similar average color.

How It Works

1. An input image is resized to a manageable resolution
2. Frames are extracted from an input video
3. The average color of each frame is calculated
4. Each pixel in the resized image is matched with a video frame of similar color
5. The final collage is created by placing the matching frames in a grid

Installation

Prerequisites

• Python 3.6+
• ffmpeg (must be installed and available in your PATH)

Setup

# Clone the repository
git clone https://github.com/natyavidhan/pixi-collage.git
cd pixi-collage

# Install required Python packages
pip install pillow opencv-python numpy scikit-learn

Usage

1. Place your input image as input.jpg in the project directory
2. Place your input video as input.mp4 in the project directory
3. Run the script:

python main.py

The output collage will be saved as final_collage.jpg in the project directory.

Examples

Original Image

Resulting Collage

Close-up Detail

Advanced Usage

You can modify the following parameters in the main() function:

# Resize the input image to a different size
resize(input_path="input.jpg", output_path="resized_image.jpg", max_size=256)

# Extract frames with different parameters
extract(input_path="input.mp4", output_dir="frames", frame_rate="1/2", size=64)

# Change the number of neighboring frames to consider for color matching
build_collage(img, filenames, color_tree, k_neighbors=5)

Parameter Details

• max_size: Maximum dimension (width or height) of the resized input image. Higher values create larger collages.
• frame_rate: How many frames to extract per second of video (e.g., "1/2" means one frame every two seconds).
• size: The size of each tile in pixels.
• k_neighbors: The number of color-similar frames to randomly choose from for each pixel.

Performance Tips

• For faster processing, use a smaller max_size for your input image.
• The frames directory is reused if it exists, so you only need to extract frames once.
• Frame loading is cached, which improves performance for repeated frames.