Remove impulse noise using randomized harmonic propagation

An image restoration algorithm named Randomized Harmonic Propagation (RHP) is proposed to remove extensive salt-and-pepper noise. The values of uncorrupted pixels are propagated to the corrupted ones iteratively under the constraint of locally self-similarity of the natural image. Compared with traditional methods, the information of all the uncorrupted pixels can be used to estimate the value of each corrupted counterpart. Extensive simulation results show that the proposed algorithm is effective and comparable with many existing, well-accepted methods in impulse noise restoration.