This paper deals with the problem of adaptive digital transmission systems for lossless reconstruction. A new system, based on the principle of non-uniform transmission, is proposed. It uses a recently proposed algorithm for adaptive stable identification and robust reconstruction of AR processes subject to missing data. This algorithm offers at the same time an unbiased estimation of the model's parameters and an optimal reconstruction in the least mean square sense. It is an extension of the RLSL algorithm to the case of missing observations combined with a Kalman filter for the prediction. This algorithm has been extended to 2D signals. The proposed method has been applied for lossless image compression. It has shown an improvement in bit rate transmission compared to the JPEG2000 standard.