使用OpenCL應用於影像對位

Applications of OpenCL in Image Registration

Chih-Yang Lin ； Wei-Sheng Lin

《Journal of Information Technology and Applications(資訊科技與應用期刊)》 8卷1期 (2014/03) Pp. 10-14

https://doi.org/10.6302/JITA.201403_8(1).0002

摘要

本文提出一個使用開放式計算語言（Open Computing Language, OpenCL）的平行處理能力加速影像對位（Image Registration）的方法，可以廣泛地應用在生物或醫學影像上，例如同步輻射（SynchrotronRadiation）影像在拍攝的過程當中，常因機器的震動而導致拍攝出來的圖片會產生位移的現象，如果以單線程程序來進行影像對位，會因計算量龐大，而消耗不少時間，因此我們使用開放式計算語言的平行處理特性來加速對位計算。對位方法則是使用對位效果不錯的共同資訊（Mutual Information）量去做影像對位，但使用單線程計算所花費的時間較長，因此我們把共同資訊量的計算使用GPU做平行處理，以達到增加程式執行效率。實驗結果顯示在開放式計算語言環境下實做影像對位，效能可提升約35%以上。

關鍵字

影像對位 ； 共同資訊量 ； 開放計算語言

並列摘要

This paper proposed a method which accelerate the speed of image registration by using OpenCL with parallel processing capacity. The method can be apply in Biomedical Images widely. For example, when synchrotron radiation image is shooting that would be usually destructed by machine vibration and this destruction could result in image displacement. If we use single-threaded program to registration images could waste much time due to huge amount of calculation. So that we use OpenCL to accelerate registration calculation. Mutual information is a good method widely used in image registration, so that we use the mutual information to register images. Single-threaded program would cost much time when it is running. So that we use GPU to calculate mutual information, it will raise program execution efficiency. The experimental result shows that the efficiency could raise above 35% when we do image registration by using OpenCL.

並列關鍵字

image registration ； mutual information ； open computing language