Computer Engineering and Applications ›› 2010, Vol. 46 ›› Issue (20): 56-59.DOI: 10.3778/j.issn.1002-8331.2010.20.016

• 研发、设计、测试 • Previous Articles     Next Articles

SIFT algorithm parallel implementation and application

HAN Long,GUO Li,LI Yu-yun   

  1. Department of Electronic Science and Technology,University of Science and Technology of China,Hefei 230027,China
  • Received:2009-06-16 Revised:2009-08-03 Online:2010-07-11 Published:2010-07-11
  • Contact: HAN Long


韩 龙,郭 立,李玉云   

  1. 中国科学技术大学 电子科学与技术系,合肥 230027
  • 通讯作者: 韩 龙

Abstract: Scale Invariant Feature Transform(SIFT) has been widely applied to many computer vision problems.But it has higher complexity.To tackle this challenging situation,an improved SIFT approach is proposed to generate panorama.Based on the multi-core system platform,this approach first uses parallel optimization to implement SIFT algorithm.And then it estimates the perspective motion model to align the adjacent views by SIFT features.Through the blending,the video panorama can be synthesized.The experimental results show that the proposed method can real-time deal with video mosaic,and also effectively synthesize different views with misregistration or parallax.

Key words: Scale Invariant Feature Transform(SIFT) algorithm, parallel optimization, image registration, video mosaic

摘要: SIFT算法广泛应用于计算机视觉等领域,而该算法具有较高复杂度,为此提出了一种基于改进SIFT特征提取的视频全景合成方法。该方法采用了多核系统平台并行优化来实现SIFT算法,首先对相邻视点的特征进行配准,然后对重合区域进行双线性融合,减少了视差的影响,同时克服了SIFT算法高复杂度问题。实验结果证明,SIFT算法的并行实现能够实时处理视频拼接,并在大角度旋转情况下仍能较好地合成视图。

关键词: 尺度不变特征变换(SIFT)算法, 并行优化, 图像配准, 视频拼接

CLC Number: