Neural Image Representations for Multi-image Fusion and Layer Separation

AbstractWe propose a framework for aligning and fusing multiple images into single view using neural image representations (NIRs), also known as implicit or coordinate-based representations. Our targets burst that exhibit camera ego motion potential changes in the scene. We describe different strategies alignment depending on nature of scene motion—namely, perspective planar (i.e., homography), optical flow with minimal change, notable occlusion disocclusion. With representation, our effectively combines inputs canonical without need selecting one reference frame. demonstrate how to use this multi-frame fusion various layer separation tasks. The code results are available at https://shnnam.github.io/research/nir.KeywordsImplicit representationsCoordinate-based representationsMulti-image fusionLayer