Visual Based Prediction of Physical Characteristics for a Smart Fashion System

摘要

This paper proposes a visual based model to predict physical characteristics from the pictures of a group of customers that were taken from a smart fashion system (SFS) at a clothing retail store. This model incorporates image segmentation, object recognition and linear regression to build a prediction model. The physical characteristics that we are interested in includes people's height, weight, BMI, skin color, shoulder length, face widths at eyes level and mouth level and the distance between the eyes and chin. Eventually SFS will use these data to recommend to a customer the best suitable clothing options. An algorithm based on image processing techniques was developed to find some of these data. A linear regression model is used to predict other data which were not easy to find using the image processing techniques due to the poor illuminations in the pictures. The experiments have proved the proposed model is simple but efficient.