东南大学学报(自然科学版)
東南大學學報(自然科學版)
동남대학학보(자연과학판)
JOURNAL OF SOUTHEAST UNIVERSITY
2013年
z2期
440-445
,共6页
图像分割%Harris算子%面匹配%特征点匹配
圖像分割%Harris算子%麵匹配%特徵點匹配
도상분할%Harris산자%면필배%특정점필배
image segmentation%Harris operator%region matching%feature point matching
针对低空影像提出一种基于图像分割和Harris算子的影像匹配方法。该方法首先利用模糊C均值方法进行图像的多阈值分割,利用数学形态学方法提取独立特征面,并根据形状特性进行影像间的面匹配。然后用Harris算子提取影像的特征点,并根据已一一对应的特征面,将特征点划分为属于各个特征面的点集,因而特征点集间也形成了对应关系。最后,以特征点与邻域角点的距离和坐标方位角来描述特征点,并通过建立、比较特征向量达到影像间同名点匹配的目的。实验证明,对于数据量大、左右影像间重叠度变化大的低空影像,单纯依靠点特征几乎无法找到正确的匹配点集,而结合特征面、点共同匹配的方法匹配正确率可达100%。
針對低空影像提齣一種基于圖像分割和Harris算子的影像匹配方法。該方法首先利用模糊C均值方法進行圖像的多閾值分割,利用數學形態學方法提取獨立特徵麵,併根據形狀特性進行影像間的麵匹配。然後用Harris算子提取影像的特徵點,併根據已一一對應的特徵麵,將特徵點劃分為屬于各箇特徵麵的點集,因而特徵點集間也形成瞭對應關繫。最後,以特徵點與鄰域角點的距離和坐標方位角來描述特徵點,併通過建立、比較特徵嚮量達到影像間同名點匹配的目的。實驗證明,對于數據量大、左右影像間重疊度變化大的低空影像,單純依靠點特徵幾乎無法找到正確的匹配點集,而結閤特徵麵、點共同匹配的方法匹配正確率可達100%。
침대저공영상제출일충기우도상분할화Harris산자적영상필배방법。해방법수선이용모호C균치방법진행도상적다역치분할,이용수학형태학방법제취독립특정면,병근거형상특성진행영상간적면필배。연후용Harris산자제취영상적특정점,병근거이일일대응적특정면,장특정점화분위속우각개특정면적점집,인이특정점집간야형성료대응관계。최후,이특정점여린역각점적거리화좌표방위각래묘술특정점,병통과건립、비교특정향량체도영상간동명점필배적목적。실험증명,대우수거량대、좌우영상간중첩도변화대적저공영상,단순의고점특정궤호무법조도정학적필배점집,이결합특정면、점공동필배적방법필배정학솔가체100%。
This paper proposes a method of low-level image matching based on image segmentation and the Harris operator.First, images are segmented by using the fuzzy C-means algorithm ( FCM) , which is a multi-threshold segmentation method, and the disconnected areas are extracted by the mathematical morphology method.These regions can be matched based on their shape features.Sec-ondly, the feature points are extracted by the Harris operator.According to the one-to-one corre-spondence between the regions, feature points can be divided into some point sets which belong to the specific areas.Therefore, the corresponding relationships are established among the point sets. Finally, point features are described with distance and coordinate azimuth between the center point and the neighborhood.Thus, the homologous image points are matched by establishing and compa-ring the feature vectors.The experimental results prove that the method only based on the feature points can barely find the matching points correctly for the low-level images which have huge data and wide rangeability of the overlapping degree.However, the method which combines the feature regions and points can achieve an accuracy of 100%.