Image segmentation with a Markov random field
At the first time, the program need to learn the mean and the covariance of each class. So, it is a supervised method: we need to put one image for each class in a folder called "classe" at project root. After that, the program generates randomly a probability map. Finding the optimal map, it does a simulated annealing. In each step of simulated annealing, the program iterates on each pixel and changes the pixel class randomly. Knowing if the new class is better than the old one, it computes the old cost and the new one and it compares the difference. This new state will be accepted if new cost is better and if this state respects the Metropolis algorithm.
The project uses CMake build system:
- mkdir build && cd build
- cmake .. && make
In project root, you can find the generated executable:
./markov input output
- CMake (build tool)
- OpenCV
- Boost: filesystem
- Armadillo
