Natural Image Matting

Given an image, the code in this project can separate its foreground and background. The algorithm is derived from Levin's research[1] and I have implemented this algorithm in C++. This applicaiton requires openCV, Eigen, CMake and Suitesparse. Please make sure you have installed these libraries before building and running my application.

A pdf report is also included in report folder to help you understand how this algorithm works.

Before Running

1. Install OpenCV

It can be much easier to install OpenCV on macOS with Homebrew.

1. If you do not have Homebrew, please install it.
2. Run the following commands to install OpenCV.

brew update
brew tap homebrew/core
brew install opencv

2. Install Eigen

This repo contains the required source files of Eigen. You do not need to install it explicitly.

3. Install Suiteparse

1. Download SuiteSparse from http://faculty.cse.tamu.edu/davis/suitesparse.html and unzip it to /imageMatting/code.

2. Build the library using the following commands (this step can take quite a while).

cd imageMatting/code/SuiteSparse;
make library;

Build and Run

1. Set the image path in src/main.cpp. Put the original image and the scribbled image in the corresponding path.
2. Build the binaries.

cd code
cd make
cmake .
make

3. The application can now run on the image provided in main.cpp.

./imageMatting

Overview of My Implementation

This section explains the function of each file in my implementation. These files are in code/src folder. For the details and the workflow of this algorithm, please refer to the report in report folder.

1. main.cpp

The main function to run the ImageMatting App. The path to the source image is also assigned in this file.

2. Image.cpp

This is the abstract class of the images. It provides some methods of retrieving the properties of images.

3. ImageReader.cpp

Given the file path of an image, ImageReader will read the image and return the matrix representation of the image.

4. LaplacianCalculator.cpp

Given the matrices of the image and the scribbled image, LaplacianCalculator calculates the Laplacian of the image.

5. SparseMatrixEquationSolver.cpp

SparseMatrixEquationSolver is specifically used to solve the equation Ax = B, where A is a sparse matrix of N by N and N is the number of pixels in the image.

6. AlphaCalculator.cpp

AlphaCalculator will firstly use LaplacianCalculator to get a Laplacian matrix. Then it constructs the sparse matrix equation Ax = B.

7. MattingPerformer.cpp

MattingPerformer uses AlphaCalculator to calculate the matte alpha. Then it applies matte alpha to the original image to get matting image. The matting image is the dot product of the original image and matte alpha.

8. ImagePrinter.cpp

ImagePrinter receives the result from MattingPerformer and then prints the matting result.

Reference

[1] A. Levin D. Lischinski and Y. Weiss. A Closed Form Solution to Natural Image Matting. (http://webee.technion.ac.il/people/anat.levin/papers/Matting-Levin-Lischinski-Weiss-CVPR06.pdf)

[2] Alpha Matting Evaluation http://www.alphamatting.com/code.php