2D Iterative MAP Detection: Principles and Applications in Image Restoration

Abstract

The paper provides a theoretical framework for the two-dimensional iterative maximum a posteriori detection. This generalization is based on the concept of detection algorithms BCJR and SOVA, i.e., the classical (one-dimensional) iterative detectors used in telecommunication applications. We generalize the one-dimensional detection problem considering the spatial ISI kernel as a two-dimensional finite state machine (2D FSM) representing a network of the spatially concatenated elements. The cellular structure topology defines the design of the 2D Iterative decoding network, where each cell is a general combination-marginalization statistical element (SISO module) exchanging discrete probability density functions (information metrics) with neighboring cells. In this paper, we statistically analyse the performance of various topologies with respect to their application in the field of image restoration. The iterative detection algorithm was applied on the task of binarization of images taken from a CCD camera. The reconstruction includes suppression of the defocus caused by the lens, CCD sensor noise suppression and interpolation (demosaicing). The simulations prove that the algorithm provides satisfactory results even in the case of an input image that is under-sampled due to the Bayer mask.