In various medical and biological modalities, in particular, electron microscopy (EM), visualization of large fields of view requires acquisition of multiple overlapping frames with their subsequent reconstruction into a single panoramic image. Such reconstruction process is hampered by several factors, including different intensity scaling and imperfect localization of the acquired frames, intensity inhomogeneity within each frame, and large content variability between different frames. This poses a significant challenge not only for visualization, but also for further quantification of such panoramic images. In this work, we present a simple yet efficient data-driven algorithm that improves reconstruction of the large panoramic views using a minimal set of assumptions. More precisely, our approach fully relies on the information from the overlap regions of the neighbouring frames. Such formulation results in a linear system of equations that can be solved numerically, when supported by proper constraints. We validated our approach on a large set of highly-diverse in-house EM panoramic views and demonstrated improved performance with respect to traditional metrics as well as network training capacity.