Fast Panorama Stitching for High-Quality Panoramic Images on Mobile Phones Yingen Xiong and Kari Pulli, Member, IEEE Abstract This paper addresses the problem of creating high-resolution and high-quality panoramic images from long image sequences with very different colors and luminance in source images. A fast stitching approach is proposed for combining a set of source images into a panoramic image using little memory, and implemented on mobile phones. In this approach, color correction reduces color differences of source images and balances colors and luminance in the whole image sequence, dynamic programming finds optimal seams in overlapping areas between adjacent images and merges them together, and image blending further smoothens color transitions and hides visible seams and stitching artifacts. A sequential panorama stitching procedure constructs panoramic images. The advantages include fast processing speed using dynamic programming for optimal seam finding, reducing memory needs by using the sequential panorama stitching, and improved quality of image labeling and blending due to the use of color correction. The approach has been tested with different image sequences and it works well on both indoor and outdoor scenes1. Index Terms Mobile panorama, image stitching, fast labeling, image blending. I. INTRODUCTION A panoramic image has a wide field of view, much wider than is available on normal cameras such as those in mobile phones. By stitching together a sequence of overlapping normal images, we can create a panoramic image. Image stitching is a very important step in creating panoramas. A simple pasting of overlapping images into the final panorama produces visible seams due to changes of scene illumination and camera responses, or spatial alignment errors. The task of image stitching is to find optimal seams in overlapping areas of source images, merge them along the seams, and minimize merging artifacts. In this paper, we are creating high-resolution and high-quality panoramic images on mobile phones, so that a user can capture an image sequence of a wide range of scenes with a camera phone and see a panoramic image created immediately on the phone. A. Background Mobile phones are not only efficient communication tools, but also capable computational devices equipped with high- resolution digital cameras, high-quality color displays, and GPU hardware. Applications such as mobile augmented 1 Yingen Xiong and Kari Pulli are with Nokia Research Center, Palo Alto, CA 94304, USA (e-mail: yingen.xiongnokia.com; kari.pullinokia.com). reality, mobile local search, and mobile image matching and recognition used to only work on desktop computers, but can now run on mobile phones. Here we are building panorama applications on these devices. A panorama construction process requires a lot of computation and memory. Mobile phones only have limited resources. It is necessary to develop efficient stitching methods to fit mobile applications. B. Related Work There are two main categories of current image stitching approaches: transition smoothing and optimal seam finding. Transition smoothing approaches reduce color differences between source images to make seams invisible and remove stitching artifacts. Alpha blending 1 is a widely used simple and fast transition smoothing approach, but it cannot avoid ghosting problems caused by object motion and small spatial alignment errors. Recently, gradient domain image blending approaches 5-8 have been applied to image stitching. These algorithms can reduce color differences and smooth color transitions using gradient domain operations, producing high-quality composite images. Optimal seam finding approaches 4, 9-12 search for seams in overlapping areas along paths where differences between source images are minimal. The seams can be used to label each output image pixel with the input image that should contribute to it label which input image contributes to each output pixel. The combination of optimal seam finding and transition smoothing for image stitching has also been used in panorama applications 4, 13, and 15. Source images are combined by compositing along optimal seams. If the seams and stitching artifacts are visible, transition smoothing is applied to reduce color differences to hide the artifacts. Current panorama stitching approaches running on camera phones can be found in 13, 2, and 3. In 13, graph cut is used for finding optimal seams to merge the source images together and Poisson blending is used for smoothing color transitions. High-quality panoramic images can be obtained. However, computational and memory costs are high. In 2 and 3, source images are stitched together with a procedure including color correction, seam finding, and simple band- linear blending. The stitching process is simple. However, the quality of panoramic ima