In this section, we will quickly go through the assumptions for the 2D HMM model and the derivation of how these assumptions simplify our equations. For a more detailed derivation, please refer to the original paper.
We start by dividing the image into smaller blocks, from which we evaluate the feature vectors, and, using these feature vectors, we classify the image. In the case of a 2D HMM, we make the assumption that the feature vectors are generated by a Markov model with a state change happening once every block. We also define the relationship between the blocks based on which block comes before or after which block. A block at position (i', j') is said to come before the block at position (i, j) if ...
