Real aperture axial stereo: Solving for correspondences in blur

When there is relative motion along the optical axis between a real-aperture camera and a 3D scene, the sequence of images captured will not only be space-variantly defocused but will also exhibit pixel motion due to motion parallax. Existing single viewpoint techniques such as shape-from-focus (SFF)/depth-from-defocus (DFD) and axial stereo operate in mutually exclusive domains. SFF and DFD assume no pixel motion and use the focus and defocus information, respectively, to recover structure. Axial stereo, on the other hand, assumes a pinhole camera and uses the disparity cue to infer depth. We show that in real-aperture axial stereo, both blur and pixel motion are tightly coupled to the underlying shape of the object. We propose an algorithm which fuses the twin cues of defocus and parallax for recovering 3D structure. The effectiveness of the proposed method is validated with many examples. © 2009 Springer Berlin Heidelberg.