Abstract
High dynamic range (HDR) images contain more details of dark and bright regions, which cannot be captured by the standard cameras. Raw HDR data is represented with floating-point precision, and the commonly used lossless/near-lossless HDR image encoding formats produce files that require large storage and are not suitable for transmission. Methods have been proposed for lossy encoding using single- and dual-layer structures, but the codecs are generally more complex and often require additional metadata for decoding. We propose a dual-layer codec in which companding, which is a closed-form transformation and hence can be reversed without any additional data, is used to generate the first layer. The residual data stored in the second layer is quantized linearly and do not need metadata either for decoding. The proposed codec is computationally light, has higher accuracy, and produces smaller size files compared to the existing codecs. This has been validated through an extensive evaluation using different metrics, and the results are reported in the paper.