基于梯度塔形分解的GPU加速多源图像融合  

GPU-accelerated Image Fusion Based on Gradient Pyramid Decomposition

作　　者：潘晓辉[1] 张涛[2] 

机构地区：[1]上海政法学院现代教育技术中心,上海201701 [2]上海交通大学计算机科学与工程系,上海200240

出　　处：《微电子学与计算机》2011年第12期54-58,共5页Microelectronics & Computer

摘　　要：为了解决多分辨率图像融合算法在单纯CPU上运行效率低的问题,提出了一种用于在基于梯度塔形分解的图像融合算法中使用的新融合规则,并将整个融合算法在图形处理器上进行了设计和高效的实现,将计算密集的任务安排到图形处理器上执行.实验验证,融合图像保留了源图像中的显著特征.并且,与单纯CPU上的融合算法相比,该系统在不同尺寸的图像融合中都获得加速.展现了一种通过图形处理器提高多分辨率图像融合算法效率的方法.To address the low-efficiency issue of the traditional multi-resolution image fusion algorithms executing on CPUs,a novel image fusion rule for gradient-pyramid-decomposition based image fusion algorithms is proposed along with the design and efficient implementation of the entire fusion algorithm on Graphics Processing Units（GPUs）.The computation-intensive tasks are offloaded from the CPU to the GPU.In the experiments,the fused image preserves the salient features of the source images.Moreover,the GPU accelerated fusion system gains speedups on images of different dimensions versus the pure CPU implementation.Thus an approach to improve the efficiency of multi-resolution image fusion algorithms is presented.

关 键 词：图像融合 梯度塔形分解 图形处理器 融合规则 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]