High Performance Image Mosaicking For HJ-1 Satellites Images

Recently, China has launched two HJ-1 satellites to space for environment monitoring. Because of the large dataset transmitted from HJ-1 satellites, a more precise, faster system of remote sensing image processing is desired to meet the new requirements. Image mosaicking is a key step for many applications of HJ-1 satellites images. Based on cluster and MPI, an efficient system for large satellite data is developed to satisfy the comprehensive requirements of HJ-1 satellites image products.The efficiency and creativities of this system can be reflected by the following three aspects:(1) Good visual results. After applying pre-processing of color balance, the images are feathered to eliminate the artifacts, which makes the resulted images with good consistency among different regions.(2) Automation of mosaicking process. A new double regions growing algorithm which is based on the shape of valid overlapped areas of the input images is proposed. This algorithm adapts itself to many kinds of conditions to generate seamlines, and avoid the disadvantages of the traditional methods which need to manually draw seamlines and estimate the feathering distance. By implementing both of the two steps automatically, the whole mosaicking process is able to be fulfilled without any human interaction.(3) Parallel operations of mosaicking process. With the analysis of parallel processing, a master-slave mosaicking model based on MPI and splitting of blocks is presented. An efficient mosaicking prototype system taking the model is developed on cluster system. The software system characterizing with friendly interface, convenient operation procedure and high speedup, could preliminarily meet the needs of HJ-1 satellites images production.