The invention relates to a foot-shaped three-dimensional surface reconstruction method based on image segmentation and grid subdivision, comprising the steps of: carrying out statistics and analysis on a shoe last sample set to obtain a statistic deformation model, using a plurality of cameras to obtain images of a foot, then using the statistic model to fit the foot-shaped images, obtaining a sparse grid model, then segmenting image characteristic points from all the images, reducing the planar characteristic points into spatial points, finally using the spatial characteristic points subdivide the grid model and carrying out the subdivision iteratively, thus obtaining a foot-shaped model consistent to a target object. In the foot-shaped three-dimensional surface reconstruction method, marking points do not need to be arranged on the foot, high-accuracy laser measurement equipment is also not needed, artificial participation is not needed, only the cameras and a computer and other devices are used, and the reconstruction risk is finished by software in a full-automatic manner. A reconstruction network can be consistent to a target foot shape automatically, can capture the detail characteristics in the images automatically and can adapt to information provided by the images automatically.