Figure 1

Overview of the imaging system.

We computationally design stacks of DOEs (b) that can encode arbitrary lenses for different geometric configurations, such as relative shift or rotation. Examples include focus-tunable lenses, tunable cubic phase-plates, and axicons shown in (a) with their corresponding phase transmission functions. Given a target phase, we design two or more phase plates using complex matrix factorization (c). If more than two phase plates are used even complex optics like zoom-systems can be designed. A novel computational approach enables broadband imaging for our encoded lenses (d,e). When a scene is imaged with our diffractive encoded lens, points with different spectral distributions result in significantly different PSFs. Our reconstruction algorithms (e) jointly self-calibrates the spatially-varying, scene-dependent PSFs and recovers the latent image exploiting cross-channel statistics. Aberrations in the reconstruction are effectively removed (actual reconstruction shown here).