A Method of Optimizing Terrain Rendering Using Digital Terrain Analysis

Terrain rendering is an important issue in Geographic Information Systems and other fields. During large-scale, real-time terrain rendering, complex terrain structure and an increasing amount of data decrease the smoothness of terrain rendering. Existing rendering methods rarely use the features of terrain to optimize terrain rendering. This paper presents a method to increase rendering performance through precomputing roughness and self-occlusion information making use of GIS-based Digital Terrain Analysis. Our method is based on GPU tessellation. We use quadtrees to manage patches and take surface roughness in Digital Terrain Analysis as a factor of Levels of Detail (LOD) selection. Before rendering, we first regularly partition the terrain scene into view cells. Then, for each cell, we calculate its potential visible patch set (PVPS) using a visibility analysis algorithm. After that, A PVPS Image Pyramid is built, and each LOD level has its corresponding PVPS. The PVPS Image Pyramid is stored on a disk and is read into RAM before rendering. Based on the PVPS Image Pyramid and the viewpoint’s position, invisible terrain areas that are not culled through view frustum culling can be dynamically culled. We use Digital Elevation Model (DEM) elevation data of a square area in Henan Province to verify the effectiveness of this method. The experiments show that this method can increase the frame rate compared with other methods, especially for lower camera flight heights.

Multi-Scale Guided Attention Network for Crowd Counting

The CNN-based crowd counting method uses image pyramid and dense connection to fuse features to solve the problems of multiscale and information loss. However, these operations lead to information redundancy and confusion between crowd and background information. In this paper, we propose a multi-scale guided attention network (MGANet) to solve the above problems. Specifically, the multilayer features of the network are fused by a top-down approach to obtain multiscale information and context information. The attention mechanism is used to guide the acquired features of each layer in space and channel so that the network pays more attention to the crowd in the image, ignores irrelevant information, and further integrates to obtain the final high-quality density map. Besides, we propose a counting loss function combining SSIM Loss, MAE Loss, and MSE Loss to achieve effective network convergence. We experiment on four major datasets and obtain good results. The effectiveness of the network modules is proved by the corresponding ablation experiments. The source code is available at https://github.com/lpfworld/MGANet.

A novel image inpainting framework based on multilevel image pyramids

Image inpainting is the art of manipulating an image so that it is visually unrecognizable way. A considerable amount of research has been done in this area over the last few years. However, the state of art techniques does suffer from computational complexities and plausible results. This paper proposes a multi-level image pyramid-based image inpainting algorithm. The image inpainting algorithm starts with the coarsest level of the image pyramid and overpainting information is transferred to the subsequent levels until the bottom level gets inpainted. The search strategy used in the algorithm is based on hashing the coherent information in an image which makes the search fast and accurate. Also, the search space is constrained based on the propagated information thereby reducing the complexity of the algorithm. Compared to other inpainting methods; the proposed algorithm inpaints the target region with better plausibility and human vision conformation. Experimental results show that the proposed algorithm achieves better results as compared to other inpainting techniques.