Temporally Coherent Local Tone Mapping of HDR Video

摘要

Recent subjective studies showed that current tone mapping operatorsrneither produce disturbing temporal artifacts, or are limitedrnin their local contrast reproduction capability. We address bothrnof these issues and present an HDR video tone mapping operatorrnthat can greatly reduce the input dynamic range, while at thernsame time preserving scene details without causing significant visualrnartifacts. To achieve this, we revisit the commonly used spatialrnbase-detail layer decomposition and extend it to the temporalrndomain. We achieve high quality spatiotemporal edge-aware filteringrnefficiently by using a mathematically justified iterative approachrnthat approximates a global solution. Comparison with the state-ofthe-rnart, both qualitatively, and quantitatively through a controlledrnsubjective experiment, clearly shows our method’s advantages overrnprevious work. We present local tone mapping results on challengingrnhigh resolution scenes with complex motion and varying illumination.rnWe also demonstrate our method’s capability of preservingrnscene details at user adjustable scales, and its advantages for lowrnlight video sequences with significant camera noise.