Through the centuries, bas-reliefs have been used to portray images on medals, souvenirs, and especially coins. Previously, the production of bas-reliefs was manual, and recently automatic methods have been proposed for the design of bas-reliefs from 3D scenes and 2D images. The authors propose an automatic bas-relief generation framework from 2D photographs, reducing the computational time.

This framework uses shape-from-shading and nonphotorealistic rendering-based methods together to generate the bas-reliefs. The shape-from-shading-based method generates “bas-reliefs from single frontal photographs of human faces,” while the nonphotorealistic rendering-based method improves the final result. The authors incorporate two components in the framework. First, an offline process relates “a frontal monochrome image of a three-dimensional human face with an image of a three-dimensional bas-relief of that face.” Second, the nonphotorealistic rendering-based method improves the appearance of the bas-relief by removing the unwanted details and highlighting its important features.

For experiments, the method takes as input four human facial reference images. Twenty-six users evaluated the resulting images in relation to the reference images. The authors evaluate the method via visual inspection or a user study because of the unavailability of ground truth data. Parameters like “head position, facial expression, lighting direction, and viewing direction” improved the results with limited changes.

Incorporating nonphotorealistic rendering makes this method interesting with regard to generating bas-reliefs from 2D photographs and 3D scenes. The authors also highlight the usefulness of nonphotorealistic rendering for other applications.