Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Modern cameras can be used to capture high dynamic range (HDR) images by taking multiple low dynamic range (LDR) pictures at different exposure settings. However, DSLR cameras often store the multiple exposure pictures, which increases storage requirements proportionally to the number of exposures, while smartphones and tablets only store a tone-mapped representation of the HDR image instead of the original HDR data. In this paper, we investigate the bit rate reduction that could be achieved by storing HDR images compressed with JPEG XT, a JPEG backward-compatible HDR image compression standard, over multiple exposures JPEG. We also investigate the quality increase that could be achieved by fusing images with linear values and native sensor bit depth over pictures with sRGB gamma encoded values and 8 bits per sample. To investigate the impact of the different parameters on the performance, we perform a set of experiments with 14 contents, 4 resolutions (ranging from 2.5 megapixels (MP) to 16MP), 4 exposures configurations (2, 3, 5, and 7 exposures), 5 tone mapping operators (TMOs) used to generate the JPEG XT base layer, and 5 quality levels representative of values commonly used in smartphones and DSLR cameras. Results show that the performance is dependent on the TMO and number of exposures, with maximum average bit rate savings ranging from 10% for two exposures up to 70% for seven exposures. Additionally, quality improvements of 1.7 dB to 5.1 dB can be achieved when using the native image sensor data format.
Touradj Ebrahimi, Evgeniy Upenik, Pol Fernandez Marti
Jan Wienold, Geraldine Cai Ting Quek, Dong Hyun Kim