Laplacian Pyramid Based Convolutional Neural Network for Multi-Exposure Fusion

Multi-exposure fusion (MEF) fuses a bracket of differently exposed low dynamic range images into one high-quality image. Motivated by the classical pyramid based MEF, a Laplacian pyramid based convolutional neural network (CNN) is proposed in this paper to fuse LDR images. The network integrates the multi-resolution fusion and non-linear inference of CNN in a model, maintaining global contrast and the detail in the fusion results. With a coarse-to-fine strategy, we rebuild the results from low-resolution to high-resolution, adding details to coarse fusion results progressively. The proposed network preserves details better than traditional CNN based MEF networks.