Two-Layer Lossless Coding of HDR Images Specialized for Radiance Format
Kai Yang, Taizo Suzuki, Taichi Yoshida · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference · 2020
Two-layer coding of high dynamic range (HDR) images, which consists of a base layer for encoding the tonemapped low dynamic range (LDR) image and an enhancement layer for encoding the residual signals between LDR and HDR images, is a constructive solution as a coding framework that can simultaneously provide the LDR and HDR images. This study proposes a two-layer coding that greatly improves the compression performance while satisfying the reversibility for the Radiance format of HDR images. Speciffically, in the enhancement layer, the image is decomposed into mantissa and exponential information by utilizing the integer conversion (RGBE converter) of the Radiance format and the luminance and chrominance components of the residual signals are sparsified by the proposed exponent-adjusted mantissa prediction. The experimental results of HDR image coding show that the proposed method improved the average bitrate by 1.41 bpp compared with the best existing method thanks to the specialized scheme to the Radiance format.