Efficient Image Stitching through Mobile Offloading

Image stitching is the task of combining images with overlapping parts to one big image. It needs a sequence of complex computation steps, especially the execution on a mobile device can take long and consume a lot of energy. Mobile offloading may alleviate those problems as it aims at improving performance and saving energy when executing complex applications on mobile devices. In this paper we investigate to which extent mobile offloading may improve the performance and energy efficiency of image stitching on mobile devices. We demonstrate our approach by stitching two or four images, but the process can be easily extended to an arbitrary number of images.We study three methods to offload parts of the computation to a resourceful server and evaluate them using several metrics. For the first offloading strategy all contributing images are sent, processed and the combined image is returned. For the second strategy images are offloaded, but not all stitching steps are executed on the remote server, and a smaller XML file is returned to the mobile client. The XML file contains a homography information which is needed by the mobile device to perform the last stitching step, the combination of the images. For the third strategy the images are transformed into grey scale before being transmitted to the server and an XML file is returned. The considered metrics are the execution time, the size of data to be transmitted and the memory usage. We find that the first strategy achieves the lowest total execution time but it requires more data to be transmitted than both the other strategies.