If one wants to present noisy videos or high-dynamic range images on an ordinary low dynamic range display, a tone mapping procedure is needed. Tone mapping aims for adjusting the overall contrast between different luminance levels in the image while at the same time keeping and in some cases magnifying the contrast locally, i.e. for the details, without amplification of the image noise. For video processing the algorithms include time integration to avoid frame-to-frame flicker, and to achieve better enhancement of details and reduction of noise. A typical ingredient in tone mapping is some kind of multi-scale resolution decomposition, for which bilateral filter and wavelet decomposition are two commonly used methods.

In the Vinnova-financed project Visual Quality Measures, FCC investigates, implements and benchmarks algorithms for the enhanced display of images and videos with applications towards the automotive industry and in particular car safety. Here, the videos are typically degraded by noise as well as subject to poor or non-uniform illumination. Unprocessed and displayed on a common display with low dynamic range without the use of any kind of tone mapping, the visual quality becomes very low, making it difficult for a person to perceive the observed scene and to suitably react to possible safety threats. Data used in the project include everything from high dynamic range video to noisy night vision videos. The image processing includes multi-scale representation, tone-mapping, statistical estimation of parameters and hidden states (the “true” image), image/sensor data fusion, and modern color interpretation modeling through the CIECAM02-standard. Present standard algorithms in the literature for single image tone mapping are gamma compression, iCAM06 [4], NASA’s Retinex, and bilateral filtering [5] for tone compression. Other partners in the project are Volvo Car Corporation, Epsilon Embedded and Chalmers.

Partners: Vinnova (FFI 2009-00071), Volvo Car Corporation, FCC, Epsilon, and Chalmers.