Linear weighted sum method for predicting girths from dressed images of young males

Abstract

Two-dimensional (2D) image-based visual body measurement plays a crucial role in the intelligent development of personalized clothing customization. How to accurately predict girth sizes under the condition of dressed human bodies remains a problem to be solved. Taking young males as an example, this paper establishes a linear weighted sum prediction model for girth dimensions based on normally dressed human body images. First, a dataset of manually measured body dimensions and their corresponding 2D image data was established, and methods of contour extraction and feature point localization were used to extract the width and thickness data from the images of clothed human bodies. Second, by comparing the correlation between image width, image thickness, body weight, and girth, as well as analyzing the characteristics of human body cross-sectional shapes, a girth prediction scheme based on the linear weighted sum method is proposed. Linear regression prediction models for the girths of the chest, waist, and hip regions of the human body are established separately. Finally, a comparative analysis of the prediction results from different independent variable selection schemes and different algorithm models is conducted. The results show that the average relative error of the predicted girths for the chest, waist, and hip regions is 1.08%, which essentially meets the 1% size error requirement for clothing customization while protecting user privacy in terms of attire. This study can efficiently and accurately provide size data for users and clothing manufacturing enterprises, facilitating the digital development of virtual fitting and personalized clothing customization.