Laplacian Coordinates for Seeded Image Segmentation

This repository contains implementations of Laplacian Coordinates: Theory and Methods for Seeded Image Segmentation by Casaca et al., DOI 10.1109/TPAMI.2020.2974475. If you plan to use the code in this repository or any variant of it, please cite the paper.

---

MATLAB implementation

• Author: Wallace Casaca (wallace.coc@gmail.com)
• Version: Preliminary prototype
• License: GPLv3

Description

This code is a preliminary version of the hard-constrained Laplacian Coordinates framework (LCH) for seeded image segmentation.

The code is very simple to run and it has been implemented and tested in MATLAB 9.2 under Windows 10 (64-bit). No extra toolboxes or mex-C compilations are required to run this prototype, making it easy to use and less sensitive to OS.

Usage

• run_me: Runs the LC segmentation for the sample images in the Example_*.mat files.
• main_interactive('sample_image.png'): Provides an interactive interface for segmentation.

Note

• For any question, suggestion or bug reports, please contact wallace.coc@gmail.com.

---

C++ implementation

• Authors: Harlen Batagelo (hbatagelo@gmail.com) and João Paulo Gois (jpgois@gmail.com)
• License: GPLv3

Description

This is a cross-platform C++ implementation of the Laplacian Coordinates segmentation framework with support to:

• Soft-constrained, pixel-based Laplacian Coordinates (LC).
• Hard-constrained, pixel-based Laplacian Coordinates (LCH).
• Soft-constrained, superpixel-based Laplacian Coordinates (SPLC).
• Hard-constrained, superpixel-based Laplacian Coordinates (SPLCH).

Build instructions

Install the build tools and dependencies:

• Qt 6.

• A C++20-compliant compiler.

• CMake, Eigen 3.3, OpenCV 4.2 and SuiteSparse. On Linux these can be installed using the distribution's package manager.

  Ubuntu 20.04 and later / Debian 11 and later:

  sudo apt-get install cmake libeigen3-dev libopencv-dev libsuitesparse-dev
  

  Fedora 34 and later:

  sudo dnf install cmake eigen3-devel opencv-devel suitesparse-devel blas-devel lapack-devel
  

  macOS with Homebrew:

  brew install cmake eigen opencv suite-sparse
  

  For instructions on how to build SuiteSparse on Windows with MSVC, refer to the suitesparse-metis-for-windows project.

After installing the dependencies, build the project in Qt Creator:

• Open cpp/CMakeLists.txt in Qt Creator and configure/build the project from there.

As an alternative, build in the command line:

• Set CMAKE_PREFIX_PATH to the location where Qt 6 is installed (e.g. $HOME/Qt/6.2.2/gcc_64).

• From the cpp directory, run the following commands (assuming Release build type):

  mkdir build
  cd build
  cmake -DCMAKE_BUILD_TYPE=Release ..
  cmake --build . --config "Release"
  

If a dependency is not found during the CMake configuration, manually set the missing path using the CMake variables Eigen3_DIR, OpenCV_DIR and SuiteSparse_DIR.

Usage

lcseg input seeds output [options]

where

• input is the name of the input image file.
• seeds is the name of the image file containing the foreground and background seeds. By default, red pixels (#ff0000) correspond to foreground seeds and blue pixels (#0000ff) correspond to background seeds.
• output is the name of the output file that will contain the segmented image.

and [options] can be any combination of the following:

• --fg: Sets the color of the foreground seeds (default is #ff0000).
• --bg: Sets the color of the background seeds (default is #0000ff).
• --hard: Uses seeds as hard labeling constraints (LCH and SPLCH).
• --superpixel: Uses SLIC superpixels (SPLC and SPLCH).
• --size: Sets superpixel size (default is 100).
• --compactness: Sets superpixel compactness (default is 10).
• -b or --binary: Writes output as a binary image.
• -q or --quiet: Runs in silent mode.