What question did this study set out to answer?

This study aims to develop a comprehensive system for 3D model reconstruction and design optimization of artworks using multimodal neural networks.

June 17, 2026Open Access

Construction of 3D model reconstruction and design optimization evaluation system of artworks based on multimodal neural network and image recognition

Key Points

This study aims to develop a comprehensive system for 3D model reconstruction and design optimization of artworks using multimodal neural networks.
Constructed a multimodal 3D reconstruction framework using RGB, depth, normal, infrared, and semantic data.
Designed a multi-dimensional index system for evaluating geometric accuracy, texture quality, and structural integrity.
Utilized 500 sets of 3D samples for testing and 30 sets of artwork models for evaluation and verification.
Method C achieved an average geometric error of 0.25 mm, demonstrating superior control over error distribution.
The model effectively distinguished materials and texture complexities through feature space clustering.
Evaluation system scored highly across five indicators, highlighting its robustness in texture restoration and structural repair.

Abstract

Digital technology promotes the way of art protection, exhibition and redesign to be constantly updated. However, the traditional three-dimensional reconstruction method still has insufficient performance in complex texture, detail reduction and structural continuity. In this study, a 3D model reconstruction and design optimization evaluation system based on multi-modal neural network is proposed to meet the digital needs of artworks. The main scientific contribution of this study is to transform artwork 3D reconstruction from a single reconstruction-oriented task into an integrated framework that combines multimodal feature fusion, image-based semantic recognition, reconstruction quality assessment, and design-oriented optimization. With RGB, depth, normal, infrared and semantic data as inputs, a 3D expression framework with stable structure and fine texture is constructed. Study and construct a multi-dimensional index system, including geometric accuracy, texture quality, structural integrity, visual performance and redesign, and design an interpretable comprehensive evaluation mechanism to realize quantitative analysis of reconstruction quality. In the experiment, 500 sets of three-dimensional samples were used for geometric accuracy testing, and 30 sets of artwork models were used for system evaluation and verification. The results show that the multi-modal fusion significantly improves the geometric error control ability, and the average error of Method C is 0.25 mm, with the most concentrated error distribution and good structural continuity. Feature space clustering forms stable clusters, which proves that the model can effectively distinguish different materials and texture complexity. The design optimization evaluation system has achieved high scores in all five indicators, which shows that the model has the characteristics of repeatable editing, robust texture restoration and strong structural repair ability. Research and form an extensible reconstruction and evaluation method to provide technical support for digital preservation, virtual display and redesign of works of art.

Ask AI

Helpful

Bookmark

View Full Paper