MAP EM image reconstruction using k-nearest neighbor method in emission CT

P.J. Green (1990) and K. Lange (1990) have proposed maximum a posteriori (MAP) estimation methods using the expectation maximization (EM) algorithm. The Gibbs prior used in the algorithm penalized the difference between neighboring pixel values in a fixed-size neighborhood. Therefore, the edge property of a reconstructed image could not be preserved in an edge area. In order to preserve the edge property, the authors propose a MAP EM image reconstruction using a k-nearest neighbour method. The neighborhood is selected using the k-nearest neighbor method. Simulation results showed that reconstructed images using the proposed method had higher quality than that obtained using the conventional MAP EM method with the fixed-size neighborhood, especially in a noise-free model.<
>