A practical and economical methodology for foot shape prediction is proposed. The methodology is facilitated by the rapid improvements in digital imaging technology. Four foot shape profiles are extracted from the digital images that were taken and foot dimensions are measured as well. The profiles are compared with the feet in a shape database and those shapes that are similar to the target foot are selected to generate the predicted shape. The result of the prediction algorithms is a 3D point cloud that is able to represent the target foot shape. This study also demonstrated a method to correct the linear perspective distortion in the captured images to generate accurate foot measurements from the digital images. Two common foot alignment methods were evaluated. It is shown that the p...[ Read more ]

A practical and economical methodology for foot shape prediction is proposed. The methodology is facilitated by the rapid improvements in digital imaging technology. Four foot shape profiles are extracted from the digital images that were taken and foot dimensions are measured as well. The profiles are compared with the feet in a shape database and those shapes that are similar to the target foot are selected to generate the predicted shape. The result of the prediction algorithms is a 3D point cloud that is able to represent the target foot shape. This study also demonstrated a method to correct the linear perspective distortion in the captured images to generate accurate foot measurements from the digital images. Two common foot alignment methods were evaluated. It is shown that the principal axis alignment method demonstrates more robust performance. An experiment with a total of 50 Hong Kong and Mainland Chinese subjects was conducted to validate the prediction error of the proposed methodology. The shape differences between the target foot shape obtained from a 3D laser scanner and the predicted foot shape was analyzed. The result shows on average a prediction error of 1.05mm (SD: 1mm). The toe and malleolus regions had the higher errors, primarily due to the smoothing operation on the reference foot shape and the postural differences during image capture and laser scanning.