Prediction of men's dress shirt pattern from 3D body measurements

Abstract

Nowadays, there are a number of developments of advanced three-dimensional body scanners in the world, all aiming at achieving better fit for diverse consumers. However, a good fit for individual consumers cannot be achieved without establishing the accurate relationship between the fit design and the body anthropometrical data. Traditional methods of pattern drafting are based on basic blocks with relevant formulae and tailor's experience. These methods of pattern drafting have not been validated experimentally and the patterns from these methods may only fit a limited group of people. In order to automate and improve the accuracy of the pattern drafting process, geometrical models have in the past developed and built in the CAD system to convert three-dimensional body measurements into the two-dimensional pattern. Such work was very successful for close-fitting apparel, but may have inherent difficulties for loose-fitting garments, which involve the complex draping of the fabric. In this study, 59 male subjects, representing the Chinese male population in Hong Kong were invited to take part in this experiment. After comparing the shirt pattern parameters calculated from the traditional drafting formulae with the actual measurements of the fitting shirts made by the tailor, considerable differences were found. One possible reason for the inadequacy of these formulae is that, each pattern parameter may relate with multiple body measurements. To address the above problem, multiple linear regression (MLR) analysis was applied to identify the important body parameters for each pattern parameter and establish the underlying relationship between body measurements and pattern parameters. Although, the results showed that the prediction of shirt pattern based on the MLR model has much improved accuracy compared with the traditional pattern drafting formulae, the prediction is still not accurate due to the possibly non-linear relationships between the pattern parameters and body measurements. Therefore, Artificial Neural Network (ANN) has been applied to establish shirt pattern prediction model from 3D body measurements. ANN has also been applied to establish a model for the prediction of the fitting perception of men's shirts from 3D body measurements and pattern parameter, which can be used by the pattern designers to evaluate the pattern in terms of fitting to the target customer before an actual garment sample needs to be produced. Furthermore, an Artificial Neural Network model was also established to predict the pattern parameters of men's dress shirts from both the 3D body measurements and fitting requirements. The validity and accuracy ANN models established in this study was further evaluated. It was found that ANN model is not only improving the accuracy of pattern prediction, but it is also providing good fit to the wearers. The study improved our understanding of the interrelationship between the 3D body measurements, fit perceptions and pattern design. The developed models can be coded into a computer program for automatic pattern generation, which is much desired for the implementation of mass customization. Although the present study is limited to men's dress shirts, the original methodology can be applied to the other types of garments in future studies.