A model for predicting the similarity between real and virtual T-shirts based on sensory evaluation and image quantitative index evaluation

Abstract

Reported herein is a model for evaluating and predicting the similarity between real clothing and clothing simulated by the virtual fitting software Vidya, the aim being to estimate accurately the similarity between real and virtual clothing. Firstly, we make real T-shirts and virtual T-shirts. According to Kawabata Evaluation System test parameters, 24 kinds of digital fabrics were constructed, and 96 virtual T-shirts were modeled from these digital fabrics. Then, 24 real T-shirts were made from the corresponding fabric. After that, the similarity evaluation indexes of virtual T-shirts and real T-shirts were extracted. Expert evaluation and digital image processing were used to evaluate the visual sense of the images and extract the indices of the evaluation-site images. Finally, the principal component analysis (PCA)-backpropagation (BP) evaluation model was established. Based on PCA, effective evaluation indices for real and virtual T-shirts were selected to construct an evaluation index system of clothing modeling and a BP neural network evaluation model. After repeatedly debugging the number of neurons, the model was optimized with four hidden-layer neurons, and the overall correlation coefficient R of the model reached 0.91075. In this paper, an evaluation index system for real and virtual T-shirts was constructed and, based on the indices, a rapid method for evaluating the virtual fitting modeling effect based on the PCA-BP model was established, which also provides a reference for the whole process of modeling evaluation of other virtual fitting software.