Images Based On Multi-core Architecture System With Quasi-parallel Calculation Method

Image registration is to find a suitable deformation, which is the process of aligning two or more images of the same scene taken at different times, from different viewpoints, and/or by different sensors. It is a crucial step during image processing. It is be widely used in medical imaging, remote sensing, computer vision and pattern recognition.The process of Image registration is to compute the pixel matrices between source image and target image. As the image size increases, more time consuming is needed. It is difficult to complete the process of registration of large size image in a limited time by traditional sequential algorithm. Nowadays, the multi-core processor is developed rapidly, we do research on parallel image registration algorithm based on multi-core platform has lot of benefits, not only makes full use of the computation resources, but also satisfies the requirement of real-time image processing.In this thesis, a parallel image registration algorithm based on multi-core platform is discussed. First, the backgrounds of image registration and development history of multi-core processor are discussed. After constructing the PDE model for registration, adopt the AOS computation formation and adaptive parameter computing method. For full utilize the multi-core computation resources and accelerate processing speed, we implement a multi-threads parallel registration algorithm by adopting the technology of data decomposition and using OpenMP parallel programming interface. And on the basis of above discussion, we discuss another parallel method on cluster using multi- processes and multi-threads. Some experimental results show that the discussed parallel methods can make full use of the computation resources, produce the large size parallel image registration quickly and save a lot of computing time.