基于不变特征和提升小波的矿井图像拼接  

Mine Image Stitching Based on Invariant Feature and Lifting Wavelet

作　　者：厉丹[1,2] 肖理庆[1] 田隽[1] 孙金萍[1] 

机构地区：[1]徐州工程学院江苏省大型工程装备检测与控制重点建设实验室,江苏徐州221008 [2]徐州市机器视觉应用工程技术研究中心,江苏徐州221008

出　　处：《小型微型计算机系统》2014年第7期1671-1675,共5页Journal of Chinese Computer Systems

基　　金：江苏省产学研联合创新资金项目(BY2009114)资助;江苏省高校自然科学研究项目(11KJD120003)资助;徐州市科技计划项目(XM12B078)资助

摘　　要：针对矿井视频中普遍存在的分辨率低、模糊、噪声大、光线变化等现象提出一种适合井下复杂环境的拼接算法.算法利用互功率谱相位信息实现图像快速粗匹配,确定重叠区域后引入样本主成分分析,用降维后的SIFT算法提取不变特征,通过改进的RANSAC结合L-M非线性优化迭代求精投影矩阵,最后根据图像数据相关性制定提升小波多分辨率融合规则,较好实现无缝拼接.实验证明新算法对复杂环境具有一定鲁棒性,拼接准确、平滑,有助于更好把握全局信息.Proposed a new stitching algorithm for the widespread existence of low resolution,blur,large noise,lighting changes characteristics in complex underground environment. Firstly,cross-power spectrum phase information is used to coarse matching fast and overlapping regions can be found. Invariant features are extracted by using the dimensional reduced SIFT algorithm-sample principal component analysis. RANSAC combined with iterative L-M nonlinear optimization algorithm can improve projection matrix. Finally,using lifting wavelet transform for multi-resolution fusion and using image date correlation to make fusion rule for seamless stitching.Experiments show that the new algorithm has a robust characteristic and stitch accurately and smoothly,it is better to grasp the overall message.

关 键 词：拼接 相位相关 样本主成分 多分辨率 提升小波变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]