CAJ | 학술논문

依据已设计完成的基于同心球透镜的四镜头多探测器阵列拼接成像系统，对该系统图像拼接配准过程所采用的特征检测提取、特征向量匹配与筛选、空间变换模型参数估计等算法进行了研究。首先，采用Fast-Hessian检测子提取参考图像和待配准图像的特征点，并生成加速鲁棒特征（ SURF）描述向量。接着，采用快速近似最近邻（ FANN）逼近搜索算法获得初始的匹配点对，并对匹配点对特征向量的欧式距离进行排序。然后，参照成像系统光学设计参数设定合理的阈值，筛选并保留下较好的匹配点对。最后，提出了一种改进的渐进式抽样一致性（ IPROSAC）算法对空间变换矩阵模型进行参数估计，从而得到参考图像与待配准图像的空间几何变换关系。实验结果表明：该算法对图像尺寸、旋转和光照变化都具有一定的不变性，特征匹配时间为0.542 s，配准变换时间0.031 s，配准误差精度小于0.1 pixel，可以满足成像系统关于图像配准实时性和准确性的要求，具有一定的工程应用价值。
의거이설계완성적기우동심구투경적사경두다탐측기진렬병접성상계통，대해계통도상병접배준과정소채용적특정검측제취、특정향량필배여사선、공간변환모형삼수고계등산법진행료연구。수선，채용Fast-Hessian검측자제취삼고도상화대배준도상적특정점，병생성가속로봉특정（ SURF）묘술향량。접착，채용쾌속근사최근린（ FANN）핍근수색산법획득초시적필배점대，병대필배점대특정향량적구식거리진행배서。연후，삼조성상계통광학설계삼수설정합리적역치，사선병보류하교호적필배점대。최후，제출료일충개진적점진식추양일치성（ IPROSAC）산법대공간변환구진모형진행삼수고계，종이득도삼고도상여대배준도상적공간궤하변환관계。실험결과표명：해산법대도상척촌、선전화광조변화도구유일정적불변성，특정필배시간위0.542 s，배준변환시간0.031 s，배준오차정도소우0.1 pixel，가이만족성상계통관우도상배준실시성화준학성적요구，구유일정적공정응용개치。
According to the detector arrays mosaic imaging system designed with four lenses based on concen-tric spherical lens, its applied algorithms about the image registration is investigated, such as feature detection and extraction, feature vector matching and screening, spatial transformation model and parameter estimation, etc.First, the fast-hessian detection algorithm is used to find features, and generate feature vector of SURF descriptors.Second, the fast approximate nearest neighbor search algorithm is used to obtain the initial matc-hing points and to sort the Euclidean distance between feature vectors in the matching points.Then after screening the feature points, the good ones are preserved based on a reasonable threshold interval from the op-tical design parameters.Finally, the transform parameters are estimated by using the improved progressive sample consensus method and the spatial geometry transformation relationship is obtained about the reference image and registration image.Experimental results indicate that the algorithm has some invariance about the size, rotation and illumination changes; the feature matching time is 0.542 s, and the registration transform time is 0.031 s;the registration error precision is less than 0.1 pixel, which can meet the requirements of the imaging system about the image registration including good real-time and accuracy performance.