利用圖形處理晶片研究共同資訊量的影像對位

Abstract

近年來圖形處理晶片的計算速度非常快的增加，用圖形處理晶片加速各種計算是一個很好的選擇，而影像對位是一個很重要的影像處理工具，其主要的應用是在匹配兩張或是多張影像上，在許多大型的影像處理系統都使用著這樣的技術。在本篇論文中，我們藉由 “共同資訊量”的最大化來做影像對位的研究，並將該技術用在生物影像處理。我們的實驗資料來自果蠅腦以及生物影像的細胞，用圖形顯示卡實作2d和3d的影像對位。

In the recent decade, the computing capacities of the graphics processor units (GPUs) has a great improvement . GPU-accelerated computation is a good option for many applications. Registration is a important tool in image processing that is applied to match regions in two or more pictures. Virtually all large systems which evaluate images require the registration of images. In this thesis, image registration technique by maximization of mutual information was studied. The technique was applied to biological images. Our experimental data are fly brain and cell of biological image . We use graphics processor units to implement 2d and 3d image registration.