Parallelizing Multi-sensor image fusion using CUDA with optimization

Pixel-level image fusion would attain high image fusion quality and improve performance of image fusion for multi sensors. However, there is huge volume of data to be processed and real time fusion processing is expensive and un-scalable. Therefore, we propose a fine-grained pixel-level image fusion implementation based on CUDA (Compute Unified Device Architecture) to accelerate Multi-sensor image fusion, in which, each pixel is assigned to one CUDA thread so that pixel-level image fusion would be parallelized by many CUDA threads simultaneously. Furthermore, we employ texture memory in CUDA to improve performance for image fusion. Experimental results show that our proposed methods would improve performance remarkably than optimized CPU counterpart available. © 2012 American Scientific Publishers. All Rights Reserved.