采用SIFT的末制导图像匹配技术  

Image matching technology based on SIFT for terminal-guiding system

作　　者：邸男[1] 李桂菊[1] 魏雅娟[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《红外与激光工程》2011年第8期1589-1593,共5页Infrared and Laser Engineering

基　　金：国家高技术研究发展计划(2006AA703405F);中国科学院知识创新工程领域前沿项目

摘　　要：图像匹配技术作为末制导图像处理器的关键技术具有重要的理论意义和实用价值。近年来,SIFT算法以其优秀的匹配性能在图像匹配领域受到广泛的关注。针对传统的及现有的改进SIFT算法计算复杂、难于实际应用的问题,使用VC++对各种SIFT的改进算子进行仿真实验,对比不同算子的实验性能,提出一种更加快速稳定的SIFT算法。在尺度空间极值检测和建立特征点描述符方面进行改进,与现有方法相比,不但保持了特征的丰富性,同时大大简化了计算量,提高了运算速度。目前,算法已经植入工程硬件平台。实验表明:改进的SIFT算法能够在图像发生缩放、旋转、平移等变化,并受到噪声、光照变化和拍摄条件不同图像内容细微变化的条件下,准确识别目标;通过双DSP并行等优化方法,算法可以满足实际工程要求的小于200 ms的捕获时间指标。In the terminal-guiding system, image matching as a key technology is valuable in theory and practicality. Recent years, wide attention has been paid to SIFT for its excellent performance in the image matching area. Due to the complexity of the traditional and improved SIFT, lots of improved SIFT methods were simulated with VC＋＋ and a fast and stable SIFT with improvements was proposed in scale-space extremum detection and key point descriptor. Compared with the existing methods, the improved SIFT not only kept the character of abundance, but also reduced the computation cost considerably. Recently, the algorithm was applied in the hardware system. Experiments on lots of images show that the improved SIFT can overcome variations of the scale, rotation, translation, blur and the small distinction between the images due to the different condition to detect the object exactly. The time of the algorithm is less than 200 ms and satisfies the practical need through using two DSP parallel computing.

关 键 词：图像匹配 改进的SIFT 尺度空间 特征检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]