On global MDL-based multichannel image restoration and partitioning

In this paper, we address the problem of multichannel image partitioning and restoration, which includes simultaneous denoising and segmentation processes. We consider a global approach for multichannel image partitioning using minimum description length (MDL). The studied model includes a piecewise constant image representation with uncorrelated Gaussian noise. We review existing single- and multichannel approaches and make an extension of the MDL-based grayscale image partitioning method for the multichannel case. We discuss the algorithm's behavior with several minimization procedures and compare the presented method to state-of-the-art approaches such as Graph cuts, greedy region merging, anisotropic diffusion, and active contours in terms of convergence, speed, and accuracy, parallelizability and applicability of the proposed method.