Regularization-Based Error Concealment in Jpeg 2000 Coding Scheme

Abstract

This letter offers a solution to bit-error recovery for high-frequency subbands at high decomposition levels of JPEG 2000 bitstreams, with a novel perspective. Being inspired by the regularization approach, corrupted coefficients are estimated by imposing a smoothness constraint on the gradient of the reconstructed data. The approach can be generalized to other compression engines as well. Experimental results show that the proposed method achieves the best reported results so far, with up to 6 dB of PSNR improvement. The method does not require any parameter tuning, however, it does require numerical minimization.