Research On Image Segmentation Algorithm In 3D Reconstruction Preprocessing

Three-dimensional reconstruction based on sequence image is a convenient and effective reconstruction method.Firstly,multi-angle shooting is performed on the reconstructed target to obtain the sequence image set.Secondly,the feature extraction and stereo feature matching are performed on the acquired images to generate corresponding 3D point cloud data.Finally,a three-dimensional model is obtained by surface reconstruction of the object.The reconstruction process of this kind of algorithm is more complicated,and a large number of similar redundant information exists in the acquired sequence images,which increases the burden on the 3D reconstruction work and reduces the reconstruction effect and efficiency to some extent.Based on the virtual simulation technology of spatio-temporal related information of major historical events and its supporting platform project,this paper studies the image segmentation algorithm in the process of 3D reconstruction preprocessing.Focusing on the accurate extraction of reconstruction targets,reducing the processing of image redundancy information,and only model the target object during the reconstruction process,which can not only effectively reduce the task of the 3D reconstruction system,but also have an important significance for improving the efficiency of reconstruction work and the effect of reconstruction goals.Therefore,based on the GrabCut algorithm,this paper further studies the application of image segmentation in 3D reconstruction of ancient buildings.The main research work is as follows:(1)For the GrabCut algorithm,there is a problem that the iterative solution takes a long time and the segmentation result is under-segmented in image segmentation.A GrabCut algorithm based on non-normalized histogram is proposed.Firstly,based on the first segmentation result of GrabCut algorithm,the algorithm replaces the Gaussian mixture model iterative learning process by non-normalized histogram to calculate the pixel point belongs to the foreground or background,which reduces the time consumption.Then,a new class of nodes Bin is introduced in the composition process to compose the image to improve the segmentation precision.Finally,some pictures of MSRA1000 data set are selected for experimental verification.The experimental results show that the algorithm has obvious improvement in segmentation effect and efficiency.The advantage of improved algorithm is more obvious when segmenting background complex images.(2)Repetitive time-consuming for Gaussian mixture model modeling process,and GrabCut can only process single image,which directly increases the workload of image segmentation,affecting the efficiency and effect of 3D reconstruction.To solve the above problems,a GrabCut continuous segmentation based on mixed Gaussian model is proposed to realize continuous segmentation of similar images in sequence images.This paper first analyzes the Gaussian mixture model to judge the process that the image pixels belong to the foreground or background and the time consumed in the algorithm,and secondly,according to the characteristics of the same or similar background information of consecutive multiple pictures in the sequence image set,and when the continuous multiple images have high similarity,the parameters of their Gaussian mixture model also have extremely high similarity.There are consecutive identical or similar images in the 3D reconstruction,so the Gaussian mixture model is reused in the paper to realize the segmentation of the three-dimensional reconstructed sequence image to reduce the time overhead introduced by the repeated modeling.The algorithm is verified by the algorithm of 3D reconstruction of Kaifeng ancient buildings in the project.The experimental results show that the improved algorithm reduces the workload of image segmentation and achieves the segmentation of target precision.The efficiency and effect of 3D reconstruction are obviously improved,and the expected effect is achieved.