Large-scale Remote Sensing Image Texture Compression And Spatial Data Integration

Spatial data processing is an important field in virtual reality. The most important part in spatial data processing is texture data compression. The texture data is almost the biggest part of spatial data. If it could be compressed efficiently, both the spatial data size would be greatly reduced and the workload of the rendering engine would be lightened. Epitome-based texture compression algorithm extracts repetitive patterns and structures in a texture, and packs them into an epitome map. It uses a transform map to restore the texture.In this paper, we propose a framework to accelerate the epitome-based texture compression algorithm. The framework employs KD-Tree to accelerate image pattern matching. It uses a new feature descriptor to perform nearest neighbor searching. The algorithm speed was improved significantly based on this framework.We propose a large-scale texture compression system based upon the accelerated algorithm. We introduced the concept of public epitome database. This database stores the common pattern of images. Working with public epitome database, the system could compress large number of images. We also implement a GPU-based real-time texture decompression rendering system.This paper also presents a resource integration pipeline for large-scale spatial data, which was based on our Visionix Virtual Reality platform. It can import various types of spatial data, including remote sensing images and building models. With these models'spatial information, the resource integration pipeline could integrate them into a unified large scene.