红外与激光工程
紅外與激光工程
홍외여격광공정
INFRARED AND LASER ENGINEERING
2013年
12期
3496-3501
,共6页
局部不变特征%仿射变换%SIFT%图像匹配
跼部不變特徵%倣射變換%SIFT%圖像匹配
국부불변특정%방사변환%SIFT%도상필배
local invariant feature%affine transformation%SIFT%image matching
SIFT(Scale Invariant Feature Transform)特征由于具有旋转、平移和尺度不变性在图像匹配中得到了广泛的应用。但直接运用SIFT特征进行匹配,存在两个问题:易受匹配参数的影响,出现较多的错漏匹配现象;只适用于相似变换情况下的图像匹配,对于高维的仿射变换情况则难以奏效,而在实际图像匹配中这种情况更为常见。针对以上问题,提出了一种空间变换迭代的SIFT特征图像匹配方法。把SIFT特征点集匹配转化为SIFT特征向量与点集的几何分布信息相关的函数最优化求解问题,通过在确定性退火框架下,迭代求解空间仿射变换与点集匹配对应关系,最终得到最优的SIFT特征点匹配关系。仿真实验表明:在较大仿射变换情况下该方法仍能实现图像SIFT特征点集的正确匹配。
SIFT(Scale Invariant Feature Transform)特徵由于具有鏇轉、平移和呎度不變性在圖像匹配中得到瞭廣汎的應用。但直接運用SIFT特徵進行匹配,存在兩箇問題:易受匹配參數的影響,齣現較多的錯漏匹配現象;隻適用于相似變換情況下的圖像匹配,對于高維的倣射變換情況則難以奏效,而在實際圖像匹配中這種情況更為常見。針對以上問題,提齣瞭一種空間變換迭代的SIFT特徵圖像匹配方法。把SIFT特徵點集匹配轉化為SIFT特徵嚮量與點集的幾何分佈信息相關的函數最優化求解問題,通過在確定性退火框架下,迭代求解空間倣射變換與點集匹配對應關繫,最終得到最優的SIFT特徵點匹配關繫。倣真實驗錶明:在較大倣射變換情況下該方法仍能實現圖像SIFT特徵點集的正確匹配。
SIFT(Scale Invariant Feature Transform)특정유우구유선전、평이화척도불변성재도상필배중득도료엄범적응용。단직접운용SIFT특정진행필배,존재량개문제:역수필배삼수적영향,출현교다적착루필배현상;지괄용우상사변환정황하적도상필배,대우고유적방사변환정황칙난이주효,이재실제도상필배중저충정황경위상견。침대이상문제,제출료일충공간변환질대적SIFT특정도상필배방법。파SIFT특정점집필배전화위SIFT특정향량여점집적궤하분포신식상관적함수최우화구해문제,통과재학정성퇴화광가하,질대구해공간방사변환여점집필배대응관계,최종득도최우적SIFT특정점필배관계。방진실험표명:재교대방사변환정황하해방법잉능실현도상SIFT특정점집적정학필배。
For the rotation, translation, scale invariant properties of SIFT (Scale Invariant Feature Transform) feature, it has been widely applied in imaging matching. But there are two defects of using SIFT while matching. Firstly, the matching performance is directly affected by the matching parameters, and there is always mismatching and error matching existed. Secondly, it only fits for matching under similarity transformation, while at the affine transformation situation it fails. In this paper, a novel iterative matching algorithm based on transformation estimation was proposed. The SIFT matching problem was turned into an optimization problem about SIFT feature vector and the geometry distribution of the point sets. By searching for the affine transformation and correspondences under the iterative deterministic annealing frame, the algorithm got the optimal matching result of SIFT point sets. Experiment results show that even at large affine transformation, the algorithm can still get the right matching results.