Finding regions of interest in document images by planar HMM

We present a new stochastic approach based on planar hidden Markov models (PHMM) for finding regions of interest (ROI) in document images. The main advantage of the proposed approach is that no explicit rules characterizing the ROIs have to be defined Instead, the PHMM learns to find the ROIs based on training data which show examples of document images together with the ROIs. The method has been tested on the task of finding the legal amount in real check images.