Freeform surface inference from sketches via neural networks

A novel approach of constructing a freeform surface from 2D planar sketches is proposed. A multilayer perceptron (MLP) neural network was employed to infer 3D freeform surfaces from 2D freehand curves. Planar boundary strokes of a surface patch produced the training set. The neural network (ANN) output mapped 2D points of the sketch curves onto 3D control points of the surface boundary. Internal points were interpolated to create the final surface. Experimentation determined the optimal parameters and ANN architecture. The methodology was applied to synthetic and realistic data. The results demonstrate successful 3D surface inference from planar sketches.