This paper adopts the matrix reconstruction method to analyze the digital image of urban landscape information in-depth and uses it to study and design the visual communication of its construction. Based on the special structure and properties of Toeplitz matrices, a stepwise structured Augmented Lagrange Multiplier (SALM) algorithm for Toeplitz matrix filling is proposed by introducing structured operators. The main idea of the algorithm is to structure the iteration matrix at each step, that is, to reassign the elements on each diagonal of the matrix by the operator. In this way, the approximation matrix always maintains the Toeplitz structure during the iterative process, and the fast singular value decomposition of the Toeplitz. A matrix can be exploited, thus saving time. The convergence theory of the new algorithm is further discussed. A lightweight progressive feature fusion module is designed to improve network learning efficiency, which consists of two components: progressive local connectivity and feature attention. Specifically, progressive local connectivity extracts multilevel features for fusion by layer-by-layer separation and local splicing. Feature attention evaluates the importance of features using both channel and spatial dual attention modules. The Oculus Rift virtual reality device system is used, and the OSG graphics rendering engine is used as the basis for the scene data transfer of the overall 3D cityscape towards the virtual reality headset display device system. With the support of the Oculus SDK, the secondary rendering of the overall 3D cityscape in the immersive virtual reality module is carried out, and the corresponding OSG camera browsing interface is constructed to realize the two-eye immersive virtual display of the 3D cityscape in a virtual reality headset display device, assisting in connecting the virtual cityscape to reality.