An Information-Theoretic Approach to Multi-Exposure Fusion via Statistical Filtering using Local Entropy
Johannes Herwig, Josef Pauli · 2010
An adaptive and parameter-free image fusion method for multiple exposures of a static scene captured by a stationary camera is described. The notion of a statistical convolution operator is discussed and convolution by entropy is introduced. Images are fused by weighting pixels with the amount of information present in their local surroundings. The proposed fusion approach is solely based on non-structural histogram statistics. Its purely information-theoretic view contrasts the phyiscally-based photometric calibration method of high dynamic range (HDR) imaging.