Research And Application Of Point Cloud Model Completion Algorithm Based On Local And Global Feature

The ancestors of Chinese civilization have created a unique and brilliant culture in their long history,leaving many valuable historical relics for future generations.The Lotus Tan Two Buddha Temple,a rare Buddhist Zen statue gathering point in the country,has witnessed the history since the late Tang Dynasty.The temple has thousands of Buddha statues,which are beautifully shaped and feature the fusion of Chinese and Western arts.Due to natural factors,a large number of Buddha statues have suffered different degrees of weathering and damage,and there is an urgent need to restore and digitally preserve the existing statues.Therefore,in order to preserve the three-dimensional data of the statues in the temple on the one hand,and not to cause secondary damage to the statues in the process of researching them on the other hand,the existing computer technology can be used to digitally scan and store the statues and reproduce the possible complete appearance of the statues through the three-dimensional model complementation algorithm,so as to provide a reference for the conservation and research of cultural relics.This can provide a reference for conservation and research.Most of the current model completion algorithms are aimed at missing data due to objective factors such as occlusion during the scanning process,and can use the model itself to infer the data information to obtain the completion results,or can use deep learning to complete.For the parts of the Buddha statue damaged by natural factors,it is difficult to use its own information for inference to complete,and there is no applicable heritage data set for deep learning-based completion.For the case of damaged parts rather than missing data,this paper proposes a sample-completion method based on multi-scale feature extraction to complete the construction of damaged parts of Buddha statues,which uses the complete sample model to complete the defective model and belongs to the practical application of non-rigid alignment.Regarding the calculation of correspondence,this paper proposes a correspondence calculation method based on multi-scale feature detection,which constructs correspondence between different Buddha models,i.e.,extracts similar parts of the Buddha as a whole by local and global feature fusion,performs feature point selection at these parts,constructs sparse correspondence based on the selected feature points,and performs rigid alignment of the target model based on this sparse correspondence The sparse correspondence is updated based on the rigid alignment of the target and source models.Finally,by matching the nearest neighbors of the correspondence,a more reliable overall dense correspondence is obtained.Regarding the calculation of non-rigid alignment,a complementary algorithm of non-rigid alignment based on the deformation framework is proposed,which defines different weights for the artificially marked defective and non-defective regions,and can preserve the local geometric features of the source model,so as to achieve the effect of complementation.