Abstract:

Objective In the process of performing virtual display-oriented reconstruction based on 3-D scanned garments, pattern generation is crucial for ensuring the accuracy of the 3-D reconstruction model, and it is influenced by the flattening and curve-fitting methods. Therefore, the selection of appropriate methods is pivotal in enhancing the accuracy of the generated 2-D patterns, which in turn enables the reconstruction of highly faithful 3-D garment models.

Method The study was focused on three types of garments, i.e., the suit jacket, trousers, and skirt. 2-D patterns were generated through a digital process, including scanning, cutting, flattening, and pattern modification. In the flattening process, the optimal flattening algorithm was selected based on the comparison of changes in length and area during the flattening process. Subsequently, comparative experiments were conducted to determine the most effective approach for curve optimization. Building upon these results, the geometric discrepancies between virtual reconstruction results and the original scanned garment results were analysed using 3-D reconstruction techniques.

Results In the context of flattening methods, the pattern shapes obtained using least squares conformal mapping, boundary-first flattening, and garment flattening(GF) were found generally similar. To quantify the differences, the flattening error was assessed using two metrics, namely, the average relative area difference and the average relative edge length difference between the flattened pieces and the original 3-D surfaces. GF was found to be the superior choice, achieving the lowest average relative area difference and edge length difference values for the right sleeve under panel and the back panel of the suit jacket, indicating minimal flattening deformation. During the pattern optimization process, comparative experiments were conducted on feature point selection and curve-fitting methods. For feature point selection, the vector angle method (sine/cosine) and the Ramer-Douglas-Peucker(R-D-P) algorithm were compared, while for curve fitting, elliptic Fourier descriptors, B-spline curves, and Bezier curves were evaluated. The combination of the R-D-P algorithm and Bezier curves yielded optimal curve-fitting results. In the armhole area of the under panel of the right sleeve of the suit jacket, as well as the armhole and hemline areas of the back panel, the feature points extracted by the R-D-P algorithm were reasonably distributed and demonstrated higher accuracy in capturing curve characteristics, effectively reflecting the curve fluctuations. Additionally, Bezier curves achieved optimal smoothness in the armhole regions of the right sleeve under panel and back panel of the suit jacket. Based on the experimental results, GF was adopted as the flattening method, with feature points extracted using the R-D-P algorithm. Bezier curves were applied to fit pattern edge curves and reconstruct interrupted curves in the pattern symmetry processing, enabling the generation of suit jacket, trousers, and skirt patterns. In the evaluation of 3-D reconstructed model, the virtual reconstruction obtained using CLO3D was overlapped with the original scanned garment model to measure the average deviation of their contours. For simpler styles, such as the trousers and the skirt, the average error between the virtual fitting result and the original scan was approximately 0.3 cm. In contrast, for more structurally complex garments, such as the suit jacket, the average error was 0.774 cm. Overall, these results indicate that the 3-D reconstructed garment models achieve high accuracy.

Conclusion This paper presents a process for generating high-precision patterns from 3-D scanned garments for virtual display. The experimental results show that it can be successfully applied to the suit jacket, trousers, and skirt, and the reconstructed models obtained through virtual sewing of these patterns exhibited high accuracy, which validates the practical applicability of this approach. Hence, the effectiveness of the proposed method for 3-D garment flattening and 2-D curve optimization was confirmed. The 2D patterns generated through the proposed workflow, can accurately reconstruct garment appearance after undergoing physical engine simulation. This approach provides high-quality virtual garment models for diverse applications, including digital preservation of cultural heritage, film and costume production, and virtual fitting scenarios.

Key words: garment virtual display, 3-D scanning, pattern generation, surface flattening, curve optimization, clothing design