Modeling a Generic Tone Mapping Operator

In recent years the problem of tone-mapping has attracted much attention and several dozens of tone mapping algorithms have been proposed. However, it is still disputable how to validate, analyze performance or quality, or simply benchmark tone mapping algorithms. The most recent operators tend to exhibit increased complexity and more sophisticated image processing algorithms, but it is not clear how much quality can be gained by this additional cost. What constitutes the improvement of the new tone mapping operators (TMOs) over the state-of-the-art? And finally, how different are tone mapping operators from each other?

In this paper we attempt to address some of these questions by modeling the processing that is performed inside a TMO. In our black-box approach, a generic tone-mapping algorithm is fit to a high-dynamic range (HDR) image and its tone-mapped low-dynamic range (LDR) rendering. We demonstrate that many tone-mapping operators, both global and local, can be satisfactorily approximated by computationally inexpensive image processing operations, such as a per-pixel tone curve, modulation transfer function and color saturation adjustment. As image above shows, the results produced by our a generic tone-mapping algorithm are often visually indistinguishable from much more expensive algorithms, such as the bilateral filter.