Automatic Field-of-View Expansion using Deep Features and Image Stitching

Abstract

Automatic photo enhancement, such field-of-view expansion, has become a challenging problem in computer graphics community. Due to the hardware limitation, image acquisition might get distracted by small field-of-view. Photo enhancement using internet photo collections has gained good performance in the past few years. However, it depends on the quality of 3D reconstruction. In this paper, we perform an automatic personal photo enhancement using the photo collection without any 3D reconstruction step. 2D global descriptor is used using NetVLAD deep architecture. Then, image stitching is applied for each similar candidate image. Experiment results show that the propose framework has promising results which could lead to further research.