VLSI design methodology for edge-preserving image reconstruction

This paper proposes a design methodology for the semi-automatic derivation of hardware dedicated to a generic class of image analysis reconstruction problems. The study will focus on the implications on the hardware of the associated estimation algorithm and of the built-in underlying minimization. A specific minimization strategy has been designed with a view to improving the efficiency of specialized hardware (in terms of clock cycle and surface area). Convergence proofs are given in the Appendix. This new nonlinear minimization has proved to be almost four times faster than the classical method used in such a context. The VLSI derivation tool presented here is based on a high-level specification of the updating rules defining the problem at hand. The complete derivation is illustrated on an edge-preserving optical flow estimator and on image restoration. (C) 2001 Academic Press.