Scott Fraser's Three way vanitas in a private collection is a carefully executed painting, commissioned through The Object Project in which several artists created paintings that included the same five objects: hand mirror, moth, ball of string, bone and drinking glass. Fraser's painting contains three plane mirrors and these provide multiple views onto the tableau.
We used image-based methods from computer vision to reconstruct the three-dimensional space and its objects based on the views in the mirrors. In brief, the method is based on so-called bundle adjustment. We hand select interest points on objects (end of a bone, tip of the mirror handle, crossing point of string, etc.) in the direct view and in the reflected views. The algorithm infers the best virtual viewing position corresponding to each mirror image to match up the corresponding sets of interest points. The resulting viewpoints give the "camera models"—the direction of view, "focal length," and so on. Then the three-dimensional positions of the points can be computed. Our method also automtically finds subtle differences between the views, for instance the height of the water in the glass as seen in the direct view and reflected in the right mirror.
Our techniques may be useful in the study of other highly realistic works to reconstruct the three-dimensional space from depicted mirror reflections. The inferred camera models may shed light on the position of the artist as he rendered different reflected images analogous to how Criminisi inferred the position of van Eyck as this artist rendered the famous convex mirror in Arnolfini portrait.