Neighborhood Size

The size of the neighborhood seems to be a modeling issue, where the mathematics may not give optimal solutions by itself. The choice of the neighborhood size typically stems from the prior knowledge, either scientific or empirical, about the physical process being modeled. For most applications in the dissertation, we have used a $9 \times 9$ pixel neighborhood. However, for certain applications, e.g., texture segmentation in Chapter 7, this neighborhood size may not work for some images. We can alleviate the sensitivity of the model to the neighborhood size by considering a multiscale adaptive-MRF model. Such a model relies on the assumption of MRFs at each level or scale of a specific multiscale image pyramid [122]. Even in such a case, some important engineering tasks persist including (a) which image decomposition to use, (b) how many levels to use in the pyramid, and (c) the size of the Markov neighborhood at each level. This dissertation does not focus on a multiscale-MRF model and such an advancement forms an important part of future work.