Human body measurement based on two-dimensional images has been widely applied in the clothing industry due to its cost and operational advantages. However, the current accuracy of human body circumference measurement is low. This article aims to propose a high-precision method for measuring human body circumference, taking bust circumference and waist circumference as examples, and based on 120 virtual simulations of human bodies, proposes a method to extract human body bust circumference size from front and side angles images. Using the feature value pixel size to calculate the trapezoid perimeter and the ellipse perimeter, and comparing them with the difference of bust circumference and waist circumference sizes, machine learning is applied to build a size prediction model, thus obtaining the values of bust circumference and waist circumference. The experimental results show that the average prediction errors of bust girth and waist girth by the proposed method are 0.26 cm and 0.24 cm, respectively, indicating good prediction performance and applicability for practical production. The proposed method effectively reduces the measurement errors of girth dimensions in image measurement and provides methods and ideas for non-contact human body measurement research.