Two layer coding of HDR images with noise bias compensation

This paper introduces a noise bias compensation (NBC) to a two layer backward compatible high dynamic range (HDR) image coding to decrease data volume of the compressed data. In this system, dynamic range of the input HDR image is reduced with tone mapping (TM) to generate a low dynamic range (LDR) image. It is encoded to generate a bit stream in the base layer. In the enhancement layer, quantization noise added by lossy coding of the LDR image is amplified by the inverse of TM. It brings about much bit rate in this layer. To cope with this problem, this paper introduces NBC to reduce variance of the noise. It compensates noise bias (NB) according to an observed pixel value. In this paper, NB is defined as the mean of the noise in the same observed pixel value from different original values. It is experimentally observed that the proposed method significantly reduces data volume in the enhancement layer at high quality lossy coding of LDR images.