The Research On Foreground Segmentation And Matching Algorithm Of Ancient Architectural Photographs

With the development of virtual reality technology and its application in various fields,the importance of 3D reconstruction technology has become increasingly prominent.Photo-based three-dimens ional reconstruction technology is an important method.It recovers the three-dimensional coordinates of a spatial point in line with the data of a single image or two images,then reconstructs it's 3D scene geometry information and accurate data.The reconstructed 3D scene can be applied not only to visualization and virtual roaming,but also to saving and recording real objects in complex large-scale scenes.The three-dimensional point cloud reconstruction method based on photo sequences can be applied in the field of cultural heritage protection,and also can be applied in aerospace,artificial intelligence,face restoration and indoor scene design.However,when a photo-based 3D reconstruction technique is used to reconstruct scenes or objects,it is inevitable that there would be problems with reconstruction quality and time efficiency.This paper aims to study the related theories and techniques of image preprocessing and feature point matching in 3D point cloud reconstruction.The following researches are carried out on the foreground segmentation algorithm and feature point matching in the image preprocessing field:1.This paper presents a foreground segmentation algorithm based on watershed segmentation.The existing foreground segmentation algorithms are mostly based on the separation of foreground and background by the GMM method,however the algorithm in this article are based on the watershed segmentation and combine with the ideas of morphological reconstruction and regional similarity measurement.First convert a color image into a binary image.Then do morphological reconstruction of the binary image to eliminate image background information.After that,the watershed algorithm is used to segment the boundary of the ancient building,then merge overly divided regions according to the principle of regional similarity measurement.Finally segment the foreground segmented image using the foreground region.The experimental simulation results show that the proposed algorithm can reduce the over-segmentation of the watershed,especially in the foreground segmentation of ancient architectural photographs it shows high efficiency and accuracy.2.This paper proposes a kind of removal mismatch method for RANSAC algorithm based on polar geometry constraint.Through the understanding and analysis of SIFT algorithm and measurement methods of feature descriptors,RANSAC algorithm and other shortcomings,it can improve the feature point matching process to removal mismatch points.Since the direction of the small number of feature descriptors generated by the SIFT algorithm is not unique,an erroneous feature point match is generated.Because of the problem of illumination and shooting angle,feature descriptors may also appear as isolated feature descriptors,and feature points cannot be matched.According to the chaotic matching of feature points,this paper first improves the measurement method of feature descriptors by replacing similarity measures based on Euclidean distances into homography matching.For RANSAC algorithm-corrected feature points matching,the correct pair of matching points is corrected by adding the contradiction of geometrical constraints to correct the wrong matching pairs.The experimental simulation results show that the proposed algorithm that performs feature point matching for the ancient building image after foreground segmentation can reduce the time for matching feature points and improve the efficiency of the algorithm.