DPCM with a recurrent neural network predictor for image compression

A new predictor for DPCM based on a recurrent neural network is proposed. Since the proposed predictor which uses a recurrent neural network called BLRNN(BiLinear Recurrent Neural Network) has shown good performance for time-series prediction problems, it is applied to DPCM for image compression with predictive coding. The performance of DPCM with BLRNN predictor is compared with the conventional DPCM with different predictors such as Linear predictor and median predictor. The results show that the proposed method gives improved results over conventional DPCM with linear predictor or median predictor in terms of PSNR of reconstructed images.