Motion-free exposure fusion based on inter-consistency and intra-consistency.

Exposure fusion often suffers from ghost artifacts, which are caused by the movement of objects when a dynamic scene is captured. In this paper, two types of consistency concepts are introduced for enforcing the guidance of a reference image for motion detection and ghost removal. Specifically, the inter-consistency, which represents the similarities of pixel intensities among different exposures, is weakened by the use of different exposure settings. Histogram matching is employed to recover the inter-consistency. Following this, pixel differences are mostly the result of changes in content caused by object movements, so motion can easily be detected. To further restrain the weights of outliers in fusion, motion detection is performed at a super-pixel level, to ensure that pixels with similar intensities and structures share similar fusion weights. This is referred to as intra-consistency. Experiments in various dynamic scenes demonstrate that the proposed algorithm can determine the motion more effectively than existing methods, and produce high quality fusion results that are free of ghost artifacts.