中文摘要：

图像融合是图像拼接中的一个重要步骤, 用于生成无缝的融合图像. 针对当前高分辨率图像融合算法速度较慢的问题, 提出一种基于多频带的快速图像融合方法. 首先通过进行2 遍距离变换生成一幅接缝图像, 并根据这幅接缝图像生成每一幅输入图像的蒙版图像, 这个过程与输入图像的次序无关, 并且具有良好的可扩展性; 其次使用游程编码对蒙版图像进行压缩编码, 生成每一幅蒙版图像的高斯金字塔, 并利用SIMD 指令集生成输入图像的拉普拉斯金字塔; 最后使用SIMD 指令集完成拉普拉斯金字塔重构, 得到一幅无缝的融合图像. 在实验中对2 组数据集进行测试, 结果表明该方法能够高效地合成高质量的融合图像. 与已有的图像融合方法相比, 文中方法运行速度更快、内存使用较少、具有更好的并行性, 更适合于目前的多核处理器架构.

英文摘要：

Image blending is an essential step in image stitching, and is often used to generate seamless blending images. However, current multi-band blending methods are much slower in processing high-resolution images. This paper presents a fast implementation of multiband blending for combining a set of registered images into a composite mosaic with no visible seams. Firstly, a unique seam image is com-puted using two-pass nearest distance transform, which is independent on the order of input images and has good scalability. Each individual mask can be extracted from this seam image quickly. Secondly, the seam image and masks are compressed using run-length encoding, and all the following mask operations are built on run-length encoding scheme. Thirdly, single instruction multiple data (SIMD) instruction set is used within Laplacian pyramids construction. Finally, the seamless blending image is generated from Laplacian pyramid collapsing. Two challenging data sets are evaluated in experiment and the proposed method can composite high quality blending images efficiently. Compared with existing image blending methods, the use of run-length encoding for masks processing leads to reduced memory requirements and a compact sto-rage of the mask data, and the use of SIMD instruction set achieves better parallelism and faster execution speed on multi-core architectures.