Automatic segmentation and splicing method of cultural relic fragment model

Abstract

The invention discloses an automatic segmentation and splicing method of a cultural relic fragment model. The method comprises steps that S1, cultural relic fragments are reconstructed into a three-dimensional mesh model; S2, triangular patches of section edge portions and section flat portions in the three-dimensional mesh model are marked, and all the marked triangular faces are further recorded; S3, a convolutional neural network model is utilized to predict the section edge portions of each three-dimensional mesh model; S4, error removal processing of the fragment model with coarse marks is performed through utilizing geometric features of the fragment model; and S5, section matching is performed through comparing section length similarity, the section pre-registration result is obtained based on the main direction mode of the section, and the ICP algorithm is further utilized to achieve accurate registration of the section to achieve splicing of a pair of sections. The method is advantaged in that the convolutional neural network is utilized to train features of the specified position, prediction is performed to obtain the fairly good segmentation effect, and accurate splicingof the cultural relic fragments is achieved.

Description

technical field [0001] The invention relates to the technical field of graphic segmentation and splicing, in particular to a method for automatic segmentation and splicing of cultural relic fragment models. Background technique [0002] In the splicing of cultural relic fragments, the cross-section of the fragments must first be found for subsequent splicing. Manually marking the fragment section requires a lot of time and effort, and there are errors. In the prior art, a three-dimensional mesh segmentation algorithm based on random walk is used to splice the fragments. This method starts from a certain point and moves to the left and right randomly. The probability of going left and right is the same, both are 1 / 2, and reach 0 Points or N points cannot be moved. Then by solving the linear equations, the probability of each point reaching the destination N point can be obtained. This idea is extended to three-dimensional grid segmentation, each triangle in the grid corres...

AI Technical Summary

Problems solved by technology

Whether it is based on the traditional 3D mesh segmentation algorithm, the 3D mesh segmentation algorithm based on spectral clustering, normal vector clustering, curvature, or MeanShift clustering, the integrity of fragment section segmentation is very poor, and redundant or missing parts may appear

Images