Error concealment using a diffusion based method

In this paper, we present a novel PDE based error concealment algorithm. We formulate the error concealment problem as a sequential optimization problem with both smoothing and orientation constraints. By introducing the orientation constraint we convert a nonlinear variational problem into a problem that is well posed and which can be solved without iterative operations. A modified orientation diffusion scheme is presented which is able to reconstruct complex orientation patterns within blocks which have been lost in an image. In the intensity reconstruction stage which follows orientation diffusion, optimization is performed based on the orientation estimates from the first stage together with the constraint of smoothness on block boundaries. We present an efficient numerical scheme which implements the method without iterations.