An image recovery algorithm based on alternating projections onto convex sets (POCS) is proposed for block-based coding. In the proposed algorithm, vectors are extracted from the surrounding neighborhood of a missing block to make a convex hull for the reference of recovery. The recovery vectors including both known and lost pixels are used for the projections to find optimal spectrum coefficients for the missing pixels. The convex hull and two other convex sets are formulated for application of POCS. Simulation results show that the proposed algorithm recovers spatial structure both at low and high frequencies.