High-quality HDR rendering technologies for emerging applications

This paper introduces high-dynamic-range (HDR) technologies that enable real-time HDR rendering for emerging imaging applications on both current and upcoming platforms. The technologies introduced in this paper are based on two key methods for real-time transformation of the photographic-quality Reinhard tone-mapping operator on emerging appliances, without compromising its quality. The first allows computing convolutions “selectively,” yielding faster computation than state-of-the-art convolution techniques while requiring a significantly lower memory footprint. The second utilizes machine learning to decrease the number of required convolutions per pixel. Both methods allow for scalable parallel implementation on commodity multicore processors and embedded processors. We also extend this implementation to 3-D and higher dimensions on massively parallel architectures for possible important applications such as medical imaging and multispectral or hyperspectral imaging applications. The inverse tone-mapping operation is equally important for rendering legacy content on new HDR displays. We used the underlying tone-mapping operator to perform inverse tone mapping using the two key methods described above. While existing techniques generally extend the dynamic range from the highlights side, our new operator performs two-sided expansion, yielding enhanced details in shades (and highlights) with generally negligible visible contrast loss. Applications of these technologies include 3-D volume rendering for the medical and seismic industries, video display on HDR television screens, and next-generation digital cameras and smart phones performing on-the-fly tone mapping.