Mantissa-Exponent-Based Tone Mapping for Wide Dynamic Range Image Sensors

摘要

The dynamic range of a scene is defined as the ratio between the maximum and minimum luminance in it. Wide dynamic range (WDR) means this ratio is so large that it exceeds the dynamic range of a traditional image sensor. Nowadays, WDR image sensors enable the capture of WDR scenes. However, the captured WDR image requires an additional tone mapping step to compress the high bit pixel of WDR image to low rate pixel so that it can be displayed on the screen. The tone mapping algorithm is mostly done in an image signal processor or with a specific software application. This brief proposes a tone mapping technique that is suitable for direct processing of the output of a WDR image sensor bitstream. The algorithm acquires statistics on the mantissa and exponent parts of the pixel value and then generates a refined histogram for tone mapping. Experiments that evaluate the image quality and hardware efficiency are carried out. The results indicate that the proposed mantissa exponent-based algorithm provides visually pleasing results and preserves details of the original WDR image better than other similar algorithms. The hardware resources' efficiency of the algorithm makes the system on chip implementation possible.