Gallery of 3D Scanning Results, Direct-Global Separation Results and Performance Gain

Here, we show several examples of 3D scanning in the presence of specularities. In each case, the 3D scanning technique used was phase-shifting of sinusoidal patterns. By clicking on each of the thumbnail images shown below, you can see the corresponding high resolution image

Conventional structured light [large exposure]

Conventional structured light [medium (best) exposure]

Conventional structured light [small exposure]

Left column shows input images for conventional SL under large (1.6s), medium (1.0s) and small (0.3s) exposure settings, respectively. Each of these sequences includes strong (saturated) specularities, dark (noisy) regions, or both. Right column shows the corresponding reconstructed shapes. At high exposures, regions with specular highlights are not reconstructed due to saturation. At low exposures, some scene regions are too dark to be reconstructed reliably.

Diffuse structured light [best exposure]

With the diffuser, the specularity is spread out and weakened in the input images. Also, the intensity of the points near the base plane is increased. Because of this, a higher quality reconstruction is achieved.

Reconstructions using best exposures for both conventional and diffuse SL, respectively. Profile views of the reconstructions are shown. Since the coin is specular, conventional SL results in depth errors (large spikes) due to specular highlights and low signal-to-noise ratio. With the same number of input images, diffuse SL produces the correct reconstruction (flat).

Alternate view-point renderings of the reconstructions using diffuse SL.

The earring is made of polished (specular) copper. Reconstructions using best exposures for both conventional and diffuse SL, respectively. Because of specularities, conventional SL results in holes. With the same number of input images, diffuse SL produces a higher quality reconstruction.

The knife blade is made of polished stainless steel. Conventional SL results in erroneous reconstructions due to specular highlights and low signal-to-noise ratio. Reconstruction using diffuse SL is nearly error-free.

This example consists of a cube on a flat plane. The projector is above the cube - the direction of projection is normal to the top face of the cube.

With conventional SL, the two vertical faces of the cube are in shadow. The cube also casts a shadow on the plane. Depth cannot be computed for the shadow regions, resulting in large holes. With a diffuser, these regions of the scene also receive illumination. Notice the projected stripes on the vertical walls and near the base of the cube in the input sequence for diffuse SL. As a result, the coverage (fraction of the reconstruction scene) of the scene with diffuse SL is significantly higher compared to conventional SL.

Here, the goal is to recover the 3D structure of the fingerprint. Without the diffuser, a large portion of the finger tip is in shadow, resulting in noisy reconstruction. With the diffuser, most of the finger receives illumination, resulting in an accurate reconstruction. Click on the thumbnails for higher resolution images.

This example consists of a pine-cone. Due to its intricate shape, a large portion is in shadow with conventional SL. This results in holes in the reconstructed shape. With diffuse SL, a larger portion of the cone receives illumination. As a result, the reconstruction is more complete.

Here, we show examples of separating direct and global components of light transport. In each case, shifting high-frequency binary stripes were used to illuminate the scene. By clicking on each of the thumbnail images shown below, you can see the corresponding high resolution image.

The grapes are translucent, resulting in subsurface scattering.

Linear diffuser preserves the high-frequency nature of the illumination, as can be seen in the input sequence of diffuse SL. Because of this, the direct and global components can be separated.

The direct-global separations with conventional and diffuse SL are qualitatively similar. There are some differences. The specularities in the direct component with conventional SL are stronger and concentrated. In comparison, the specularities in the direct component with diffuse SL have a 1D spread.

Notice the strong shadows in the direct component without the diffuser. In comparison, the shadows are softer in the direct component with the diffuser.

In order to compute the performance gain of diffuse structured light quantitatively, we consider spheres as they have all surface normals. We consider a glossy sphere for specularities and a Lambertian sphere for shadows. By clicking on each of the thumbnail images shown below, you can play videos and see the corresponding high resolution image.

Here, we illustrate the coverage of a glossy sphere. Coverage is the fraction of the sphere which can be captured reliably. For illustration, the sphere is mapped to a circle. The red portion of the sphere is too dark (intensity less than 10 grey levels) to be captured faithfully. This part is considered uncovered. The covered portion is shown in grey. For very smooth material, a very small portion of the sphere is covered without the diffuser (scattering angle of 1 degree corresponds to no diffuser). As the diffuser scattering angle increases, the coverage increases. Click on the images to play videos. Similar trend can be observed for more rough materials.

Plots of the coverage fraction for different material properties.

To quantify the performance gain in the presence of shadows, we consider the coverage of a Lambertian sphere. With conventional SL, only the top half of the sphere receives light. With the diffuser, portion of the bottom hemisphere is covered as well. As in the case of specularities, coverage increases with increasing scattering angle