Research on Constructing Image Stitching Models of Gigapixel Computational Imaging Systems

This paper addresses the issue of low accuracy in image stitching models and poor fused video quality in gigapixel computational imaging systems caused by feature-deficient areas in target scenes. An improved method is proposed, which obtains sample images from an array camera, performs region matching to determine initial positions, and calculates loss values to identify reference region pairs and classify regions as normal or abnormal. Abnormal regions are corrected through fitting, and image registration constructs the final stitching model. This approach improves model accuracy and scene-matching, enhancing fused video quality. Additionally, a fused video generation method is introduced, along with related devices, systems, and tools.