Feature Matching Based On Grid And Multi-density For Ancient Architectural Images

With the development of computer and camera,3D reconstruction of ancient architecture based on images has been widely applied in Digital Archaeology,the construction of i-city and so on.However,it has low precision and poor robustness due to the complex structure and complex acquisition environment.Feature matching is a key step of 3D reconstruction based on images,and its accuracy and efficiency directly affect the result of 3D reconstruction.In this paper,we study the theory and method in the feature matching of ancient architecture to address the mismatch and poor robustness of the feature matching.The proposed feature matching method is designed which combined the structural information among features in the ancient architecture images and used the clustering based on grid and multi-density.This main research results are as follows:1.To reduce the mismatches caused by repeated textures in feature matching of ancient architectures,a feature matching algorithm named FM＿GMC is proposed,which combined the features’ geometric information with SIFT descriptors.At first,the images with SIFT key-points can be divided into several grids,and then the density of grids is computed.Secondly,the clusters are determined according to the similarity of local grids’ density.Thirdly,the correspondence clusters are matched referring to NNDR,and then the correspondence key-points in correspondence clusters are matched also referring to the NNDR.At last,the efficiency and robustness of FM＿GMC is validated by3 D reconstruction dataset.2.To address the the poor robustness due to single measurement in feature matching,a weighted measurement is proposed,which combines the spatial information,local feature information and gray information of key-points in images.The correspondence cluster regions are acquired according to the improved FM＿GMC at first.Secondly,the edge-features are obtained in these regions by Canny and are described by SIFT.Thirdly,the Hausdorff distance,Euclidean distance and NCC are used to measure the similarity of key-points in corresponding clusters by the weighted strategy.In the end,the validity of WSM is verified by 3D reconstruction dataset.3.Based on FM＿GMC and WSM,a feature matching prototype system based on grid and multi-density for ancient architecture images(FMS)is designed and implemented.Taken the Jinci images as an example,the function of FMS includes loading images,feature extracting,clustering based on grid and multi-density and weighted similarity measurement for feature matching.The result of FMS shows that the efficiency of feature matching is effectively improved by adopting WSM and FM＿GMC,and the technical support for 3D reconstruction on ancient architecture is provided further.