ExpoBlend: Information preserving exposure blending based on normalized log-domain entropy

•We present an algorithm called ExpoBlend for blending photographs across varying exposures.•ExpoBlend weights exposures according to the local entropy of log-intensity values.•Existing high-dynamic range images or individual unmodified exposures may be blended.•A single parameter can be adjusted to preserve realism, or maximize detail.•ExpoBlend is especially adept at preserving detail within brightly lit localized regions.

In this paper, we present a solution to the problem of dynamic range compression from multiple exposures called ExpoBlend that operates in the absence of raw format images, relative or absolute exposure values, camera response functions, or known irradiance. This is achieved in relatively simplistic fashion by merging image content across provided exposures. The proposed algorithm is directed at making visible any contrast appearing across a dynamic range that exceeds display or printing capabilities through high dynamic range (HDR) compression while preserving the nature of the image structure and detail, lighting, and avoiding introducing discontinuities in illumination or image artifacts. In addition, ExpoBlend allows scaling subject to a single parameter that elicits a trade-off between the impact of illumination and fine detail in the merged result. The strategy applied appeals to an information maximization strategy wherein the local entropy evident in each exposure is computed subject to a logarithmic compression of intensities, and employs cross-exposure normalization of entropy that implies a fusion strategy based on relative entropy across exposures in combination with a soft-maximum operation.

Graphical abstractIn this paper we propose an algorithm called ExpoBlend for blending multiple exposures to capture high-dynamic-range imagery. Output may be tuned to create a photorealistic result or to maximize detail in tuning a single parameter.Figure optionsDownload full-size imageDownload high-quality image (90 K)Download as PowerPoint slide