基于多局部特征匹配的全自动图像拼接  

AN AUTOMATIC IMAGE STITCHING APPROACH BASED ON MULTIPLE LOCAL FEATURES MATCHING

作　　者：严磊[1] 汪增福[1] 

机构地区：[1]中国科学技术大学自动化系,安徽合肥230027

出　　处：《计算机应用与软件》2010年第10期5-7,54,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(60875026)

摘　　要：针对数码相机或手机等手持拍摄装置拍摄的图像,提出了一种基于SIFT和MSER互补不变特征的全自动图像拼接算法并进行了系统实现。算法将MSER检测子推广到高斯尺度空间,增强了其对尺度变化的适应性,并与SIFT检测子相结合,发挥二者在图像中空间分布上的互补性,从图像中获取更为丰富的特征,使图像间的配准更为稳健。对于存在匹配关系的多幅图像,提出了一种简单而有效的光度差异调整方法,使生成的全景图像效果更为自然。最后实验验证了该系统的有效性和实用性。In view of the images shot by handheld picture-taking devices such as digital camera or cellphone,etc.,an automatic image stitching algorithm is proposed in this paper based on SIFT and MSER complementing invariant features each other.The system implementation of the algorithm has been realised.It generalises the MSER detector to Gaussian scale-space to have enhanced its adaptability to scale changes, and by combining with the SIFT detector to allow the full play of two detectors＇ complementary in spatial distribution of the images,the algorithm acquires richer features from images and makes the registration between the images more robust.For multiple images with matching relationship,a simple but effective photometric differences adjustment algorithm is further proposed to generate the panoramic image of a scene with more natural effect.Experimental results verify the efficiency and practicality of the system at last.

关 键 词：图像拼接 图像配准 SIFT特征 MSER特征 光度差异调整 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]