Results
The 'Atlas and Candle' scene
This scene comprises of a wax candle inside a white clay pot, a marble statue of Atlas, a V-groove made of two diffuse planes, and a polyresin bust, in this depth order. The wax candle and the marble statue exhibit a significant amount of sub-surface scattering. Also, there are inter-reflections in the V-groove concavity and between the folds of the statues.
The approximate depths of the clay-pot, the marble statue and the polyresin bust are 150cm, 200cm and 270cm respectively. The depth range of the scene (~120cms) is greater than the depth of focus of the projector (~30cms). Click on images for high-resolution versions.
The approximate depths of the clay-pot, the marble statue and the polyresin bust are 150cm, 200cm and 270cm respectively. The depth range of the scene (~120cms) is greater than the depth of focus of the projector (~30cms). Click on images for high-resolution versions.
Depth Recovery under Global Illumination

Depth Map using multiple focal planes (Section 5.1). The depths are recovered correctly even in the presence of sub-surface scattering and inter-reflections. Also, since the computations are pixel-based, fine structures such as the wick of the candle are recovered. The stripe artifacts are due to the limited spatial resolution of the projector

Renderings from different view-points

Depth Map using two focal planes (Section 5.2)

Depth Map using single focal plane [23]. Incorrect depths are recovered in the presence of sub-surface scattering (candle and marble statue) and inter-relfections (inside the V-groove concavity and the folds of the statues).
Direct-Global Separation under Illumination Defocus
Separation using multiple focal planes (Section 6.1). Notice the high global component on the candle and in the V-groove concavity. On the other hand, global component is low for the clay pot.

Direct Component

Global Component
Separation using a single focal plane and depth map of the scene (Section 6.2). This separation result is more noisy as compared to the multiple focal planes algorithm. However, this algorithm requires only a single focal plane position of the projector and a depth map of the scene.

Direct Component

Global Component
Separation using a single focal plane [15]. Since the depth of the scene is larger than the depth of field of the projector, incorrect separation is produced. Here, the projector focal plane is placed behind the scene. Notice the incorrect separation in the front of the scene: low direct component and high global component on the clay-pot.

Direct Component

Global Component
'Organic' Scene

This scene exhibits significant sub-surface scattering due to the presence of a variety of organic materials: peppers, two pumpkins, the green plant, some marbles and eggs and a flower-pot, in this depth order. Additionally, there are inter-reflections between the eggs and between the peppers. Click on images for high-resolution versions.
Depth Recovery under Global Illumination

Depth Map using multiple focal planes (Section 5.1).

Depth Map using two focal planes (Section 5.2).

Depth Map using single focal plane [23]. Depths are over-estimated due to sub-surface scattering (peppers) and inter-reflections (between the eggs).
Direct-Global Separation under Illumination Defocus
Separation using multiple focal planes (Section 6.1).

Direct Component

Global Component
Separation using a single focal plane and depth map of the scene (Section 6.2).

Direct Component

Global Component
Separation using a single focal plane [15]: Different separations are produced as the projector focal plane is moved from front to the back of the scene. For each focal plane position, the computed separation is incorrect at scene points which are distant from the focal plane.
Scene with Complex BRDFs, textures and geometry

Here we demonstrate the validity of our techniques on a scene with complex BRDFs, textures and geometry. This scene consists of objects with highly specular (metal) and anisotropic (velvet) BRDFs. There are objects with complex textures as well as objects with no texture. In addition, the metal ball has a complex shape with a significant amount of occlusions and the fur teddy-bear has fine geometric structure. There is global illumination in the fur and reflections from the shiny table-top. Click on images for high-resolution versions.
Depth Recovery

Depth Map using multiple focal planes (Section 5.1). Our techniques are robust to complex BRDF's, (lack of) textures and complex geometries. Correct depth maps are computed even in the presence of severe occlusions and depth discontinuities (metal ball).

Depth Map using two focal planes (Section 5.2).

Depth Map using single focal plane [23]. Incorrect depth-map due to reflections from the shiny table-top. Notice the abrupt transition near the base of the red-cylinder and the base of the sea-shells. In contrast, notice the correct smooth transition in the depth maps computed using our techniques.
Direct-Global Separation under Illumination Defocus
Separation using multiple focal planes (Section 6.1).

Direct Component

Global Component
'Candles and Soaps' Scene

This scene comprises of different kinds of soaps and candles made of different kinds of wax. These objects exhibit varying amounts of translucency due to different material properties and pigments. All the objects are placed at the same depth. Click on images for high-resolution versions.
Depth Recovery

Depth Map using our technique (multiple focal planes). Correct depth map is computed despite different amounts of translucencies (all the objects are placed at the same depth).

Depth Map using [23]. Different amounts of translucencies introduce a bias in the recovered depths.
Direct-Global Separation

Direct Component

Global Component
'Real vs. Fake' Scene

This scene demonstrates comparisons between real and fake fruits, real and fake flowers and full-fat milk and skim milk. Can you guess which are real and which are fake? Which is full-fat milk and which is skim milk? Real and fake materials exhibit different amounts of translucencies. Also, skimmed milk is more translucent as compared to full-fat milk. Click on images for high-resolution versions.
Depth Recovery

Depth Map using multiple focal planes (Section 5.1).

Depth Map using two focal planes (Section 5.2).

Depth Map using one focal planes [23]. More translucent a material is, more the bias in the recovered depths using this method. For example, the right glass contains the more translucent skimmed milk - resulting in a larger bias. Similarly, the right lemon is real and more translucent than the fake lemon. Our techniques account for global illumination to produce the correct depths maps.
Direct-Global Separation
Separation using multiple focal planes (Section 6.1).

Direct Component
