Gallery of Separation Results
Here, we show several examples of our separation results. In each case, we mention the separation method we have used. In all cases, the scene image is the sum of the computed direct and global images. However, to bring out the details, for some of the scenes we have brightened the direct and global images by a scale factor (between 1 and 2) with respect to the scene image. By clicking on each of the thumbnail images shown below, you can see the corresponding high resolution image.
(Back to Main Project Page)
Complex Scene
Method : Checkerboard

.
Scene

.
Direct

.
Global
This example includes a variety of global illumination effects. Strong global illumination effects arise due to diffuse interreflections between the white walls, specular interreflections within the plastic cup, volumetric scattering due to the milky water in the cup, subsurface scattering in the candles and in the marble  object and translucency of the frosted glass.
 

Captured Video: This video shows the images of the scene captured by shifting a high frequency checkerboard pattern. MPEG, Quicktime

Diffuse V-Grooves : Concave and Convex
Method : Checkerboard

.
Scene

.
Direct

.
Global
This scene includes two identical V-grooves, except that the left one is concave and the right one is convex. The two sides of each groove are made of matte paper (white on left side and pink on the right). As expected, the convex groove has no global illumination while the concave one has a strong global component due to diffuse interreflections that increase toward the edge of the groove. The interreflections cause the color of one side to "bleed" into the other. Several studies have been conducted in psychophysics that show that interreflections play an important role in the perception of shape and color.
 
Building Corner
Method : Line Occluder

.
Scene

.
Direct

.
Global
This scene of a building corner lit by the sun is scanned using a stick occluder. Notice the sharp shadows cast by the sun in the direct image.  The global component on the other hand captures interreflections between the walls as well as the illumination received from the sky.
 

Captured Video: This video shows the scanning of the scene using a stick and was used to compute the direct and global images. MPEG, Quicktime

Eggs on a plate
Method : Checkerboard

.
Scene

.
Direct

.
Global
In this scene of eggs on a plate, the strong interreflections between the eggs and between the eggs and the paper are captured in the global image. The direct image shows the diffuse shading over the eggs and the sharp shadows cast by the eggs onto each other. The orange paper can be seen to glow in the global image due to diffusion of light through the paper.
 
Water tub with objects
Method : Checkerboard

.
Scene

.
Direct

.
Global
This scene includes objects in a tub filled with water. Since the water is clear it acts as a fully transparent medium that does not influence the scattering effects. It is worth noting that the computed direct image looks like a synthetic image rendered using a single-bounce rendering package such as OpenGL. All the interreflections between the sink and the objects are observed in the global image. Notice the strong interreflections at the edges and corners of the sink and the occluding boundaries of the curved objects.
 
Textured Fabric
Method : Checkerboard

.
Scene

.
Direct

.
Global
In this example of a sheet cloth with folds, the global image includes interreflections within the folds as well as translucency effects due to the fine spaces between the weaves of the fabric.
 
Curtain
Method : Mesh Occluder

.
Scene

.
Direct

.
Global
This curtain with folds was illuminated by a halogen lamp and scanned using a mesh occluder. The global image includes interreflections within the folds as well as translucency effects due to the fine spaces between the weaves of the fabric.
 

Captured Video: This video shows the scene being scanned using shadow cast by a mesh occluder. MPEG, Quicktime

Pink Carnation
Method : Checkerboard

.
Scene

.
Direct

.
Global
In the case of this flower, we see that in the direct image the shadows cast by the petals on each other are strong and the petals  themselves appear grayish and somewhat listless. As a result, the direct image looks more like that of a synthetic flower than a natural one. It is interesting to note that most of the color of the flower arises from global effects. These include the interreflection of light between the petals as well as the diffusion of light through the petals.  Both these effects cause a "sharpening" of the spectral distribution of the light. Due to this sharpening, the color of the light gets more reddish after each bounce or diffusion.
 
Wooden Blocks
Method : Checkerboard

.
Scene

.
Direct

.
Global
These colored wooden blocks have both diffuse and specular components. The computed global image shows the complex interreflections and color bleeding between the blocks. Again, similar to the V-groove scene, the interreflection is stronger near the concavities.
 
Novel Images: This video shows novel images generated by simple addition of differently weighted direct and global images. Though the resulting images are physically impossible, they exaggerate or de-emphasize the optical interactions between the blocks. MPEG, Quicktime
Mirror Ball : Failure Case
Method : Checkerboard

.
Scene

.
Direct

.
Global
This scene with a mirror sphere surrounded by three diffuse walls violates the assumption that the global component received by scene points is smooth compared to the high frequency illumination. As a result, we see a regular pattern (artifacts) due to over-estimation of the global component and under-estimation of the direct component. Looking at it in a different way, at each of these corrupted pixels there is no longer a single significant scattering event but rather two or more. Also note the checker-like pattern that appears on the mirror sphere itself in the direct and global images.
 
Photometric Stereo : Painted Bowl
Method : Line Occluder
Source 1
Source 2
Source 3
 
Regular Images
 
 
Global Images
 
 
Direct Images
 
The top row shows three images of a diffuse painted bowl captured under three different light source directions. The second and third rows show the corresponding global and direct images obtained using a stick occluder.
 
Photometric Stereo Results : Here we show the actual shape of the bowl measured manually(yellow) and the shapes computed by applying photometric stereo to the regular images(blue) and the direct images(red). As expected, the regular images produce a much shallower shape due to interreflections, while the direct images produce the correct shape.
Cup : Coin in Milky Water
Method : Checkerboard

.
Scene

.
Direct

.
Global
In this example, the global image includes the volumetric scattering of light by the milky water (referred to as "airlight" in atmospheric optics) in the cup as well as the secondary illumination of the coin by the milky water. The direct image includes the specular highlights on the coin due to direct illumination of the coin by the source. This component is attenuated by the milky water as it makes its way to the camera.
 
Kitchen Sink : Objects in Milky Water
Method : Mesh Occluder

.
Scene

.
Direct

.
Global
This scene of a kitchen sink includes objects immersed in milky water which acts as a participating medium. The global effects are mainly due to volumetric scattering by the milky medium which not only scatters light from the source but also illuminates the objects as a secondary source. It is interesting to note the faint colored glow around the objects in the global image due to light received by the milky water from the objects. Note that the direct image is devoid of any volumetric scattering and hence appears like an image of the objects taken in a clear medium.
 


Novel Image: To generate this novel image, the dark regions of the global image were used to estimate the brightness of the milk. It is assumed to be constant over the scene and is removed from the global image to obtain the radiance of the objects due to illumination by the milk. The ratios of brightnesses in the direct image and the milk illumination image were tabulated. Then, the direct images of two other objects (green fruit and orange pot) were separately captured and their milk illumination images were computed using the tabulated ratios. The direct, milk illumination and milk brightness components of these new objects were then added and the objects were inserted into the scene image. Notice how the inserted objects include not only the effects of scattering by the milk but also secondary illumination by the milk.
Curtain
Method : Mesh Occluder

.
Scene

.
Direct

.
Global
In this scene, a mannequin is placed just behind a translucent shower curtain. The mannequin is not directly lit by the source but by the light diffused through the shower curtain. The direct image only includes reflections from the shower curtain and not the mannequin. The global component reveals the mannequin behind the curtain.
 

Captured Video: This video shows the scene being scanned using shadow cast by a mesh occluder. MPEG, Quicktime

Translucent Balls
Method : Checkerboard

.
Scene

.
Direct

.
Global
The balls in this scene exhibit very strong subsurface scattering which causes them to "glow" under virtually any illumination. We see that all the subsurface scattering is captured in the global image. Perhaps, due to strong multiple scattering, the global images of the balls have very little shading, causing the balls to appear like flat discs (particularly, the green ball). On the other hand, the direct component reveals the spherical shapes of the balls and the roughness of their surfaces.
 
Peppers
Method : Checkerboard

.
Scene

.
Direct

.
Global
These peppers exhibit very strong subsurface scattering. As a result, the colors of the peppers are almost entirely captured in the global image. The direct component mainly includes specular reflections, except in the case of the green stalks that are rather more diffuse reflectors.
 

Novel Images: This video shows novel images generated by changing the hues of the peppers in the global image and adding it back to the direct image.
MPEG , Quicktime

Grapes and Cheese
Method : Checkerboard

.
Scene

.
Direct

.
Global
Similar to peppers, the colors of the grapes are mostly due to global illumination by subsurface scattering. The direct components for both sets of grapes are similar and include both specular and diffuse reflections from the surface. The cheeses in the scene also produce strong subsurface effects. The direct image also captures the specular and diffuse reflections from the plastic sheet wrapped around the cheeses. The plastic wrapper is also ‘back’ illuminated by the cheeses, an effect captured by the global image.
 



Novel Images: These images were generated by changing the hues of the grapes in the global image and adding it back to the direct image.

Bread
Method : Checkerboard

.
Scene

.
Direct

.
Global
Due to the spongy nature of the bread, most of the light passes through the surface and scatters beneath it to produce strong and complex global effects. The direct image shows the 3D texture of the bread surface as well as the albedo variations at the crust of bread and grains in the bread.
 
Candles
Method : Checkerboard

.
Scene

.
Direct

.
Global
The color of these candles is mainly due to subsurface scattering as can be seen in the global image. On the other hand, the direct image primarily includes specular reflections from the surfaces of the candles.
 

Novel Images: This video shows novel images generated by changing the hues of the candles in the global image and adding it back to the direct image.
MPEG, Quicktime

Tea Rose Leaf
Method : Checkerboard

.
Scene

.
Direct

.
Global
Although this leaf is thin, it exhibits a noticeable global component over its entire area. This is because the entire leaf has a spongy mesophyll layer, beneath the upper epidermis and palisade mesophyll layers, that are slightly translucent. The global component is particularly strong for the veins of the leaf. This is because the veins are even more translucent as they are made of vascular tissues that carry water, minerals and sap.
 
Red leaves
Method : Line Occluder

.
Scene

.
Direct

.
Global
This scene of red leaves is lit by the sun and is scanned using a thin stick occluder. The direct image has sharp and dark shadows. The global image, on the other hand, has a glowing effect due to the diffusion of light through the translucent leaves. It also includes the illumination of the leaves by the sky and surrounding buildings. Note that the green leaf beneath the red leaves (in the shadow area) does not receive any direct illumination from the sun and hence is visible only in the global image.
 

Captured Video: This video shows the scanning of the scene using a stick and was used to compute the direct and global images. MPEG, Quicktime

Marble : From BSSRDF to BRDF
Method : Checkerboard
 
Resolution Scale = 1
 
 
Resolution Scale = 1/2
 
 
Resolution Scale = 1/4
 
 
Resolution Scale = 1/6
 
Scene
Direct
Global
This example shows the separation results for a marble tile for different resolutions of imaging and illumination. For scale = 1, the resolution of the camera was set at the native resolution of 1024x768 and the separation was done with a checkerboard pattern with checkers of 4x4 projector pixels. For lower scales of 1/n, the resolution of the camera was reduced by aggregating n x n pixels and the checker size of the illumination pattern was increased to 4n x 4n. The images shown above correspond to scales 1, 1/2, 1/4 and 1/6 from top to bottom. At scale = 1, we see a slight direct component and a dominant global component. In contrast, at scale = 1/6, we see that the global component is nearly zero everywhere and the direct component is almost equal to the original image. This example shows how the BSSRDF of a translucent object dominates at fine scales, but reduces to a BRDF as one goes to coarser scales.
 
Hands of Different Nationalities
Method : Checkerboard
 
(a)
 
 
(b)
 
 
(c)
 
 
(d)
 
Scene
Direct
Global
Here, we show separation results for hands of people from four different nationalities - (a) African American female, (b) Asian Indian male, (c) Spanish male and (d) Chinese male. Notice how the direct component mainly includes the surface reflection due to oils and lipids on the skin. It also reveals the details of the micro-geometry of the skin surface.  It is also interesting to notice the difference in the micro-geometry of the different samples. Most of the color of the skin comes from subsurface scattering, as seen in the global image. In contrast to the direct component, the global  component does not reveal the roughness of the skin’s surface and only includes albedo variations due to the melanin levels within the skin.
 
Hand with Tattoo
Method : Checkerboard

.
Scene

.
Direct

.
Global
Here we show the separation results for a hand with a green tattoo. Notice that the green tattoo appears in the direct image and is very dark in the global image, indicating that the pigments of the tattoo ink are more opaque than the tissues of the skin.
 
Face
Method : Singe Image

.
Scene

.
Direct

.
Global

The separation results for this face were obtained from a single high resolution image (click here to see the image) captured with the face lit  by a high frequency stripe pattern. By assuming smoothness of surface properties, low resolution direct and global images are computed from the captured image. The resolution of the results shown above are one-fourth the resolution of the captured image. Similar to the case of hands, the specular reflections due to oils and lipids are captured in the direct image. The micro-geometry of the skin is however not visible due to reduction in resolution. The skin tone is produced by subsurface scattering and is measured in the global image.

 

Oiliness Changes: This video shows novel images generated by scaling the direct image and adding it back to the global image. MPEG, Quicktime

Skin Tone Changes: This video shows novel images generated by changing the hue of the skin in the global image and adding it back to the direct image.
MPEG , Quicktime

Different Hairs
Method : Checkerboard
 
(a)
 
 
(b)
 
 
(c)
 
 
(d)
 
Scene
Direct
Global
Separation results for four types of hairs - (a) Blonde, (b) Brown, (c) Auburn and (d) Black. Note that the direct image makes all the hairs look like dark tan hair.  The direct image reveals the complex BRDF of individual hair fibers, which is due to the nested-cone structure of each fiber. One can observe the complex structures of the specular highlights as well as rainbow scattering effects that are known to occur in hair fibers. The direct image also shows some random speckle artifacts due to the color demosaicing operation applied to the captured images. The global images for first three hair samples have the appearance of a flat texture such as that of finished wood and is almost non existent in the case of black hair. The global component arises from multiple reflections of light between the hair fibers and is seen to account for the color of the hair.
 
Blonde Hair: Bidirectional Texture Function
Method : Checkerboard

This video shows the separation results for blonde hair as a function of light source direction (varied from -30 degrees to +30 degrees in steps of 3  degrees, where 0 degrees corresponding to the viewing direction). While the global image varies slowly with respect to lighting direction, the direct component changes more drastically.  MPEG, Quicktime

 
Pebbles
Method : Checkerboard

.
Scene

.
Direct

.
Global
Here we show the separation results for the pebbles sample taken from the CURET database . Notice how the direct image look like a real image of dark pebbles. The global image includes subsurface scattering within individual pebbles as well as the complex interreflections between neighboring pebbles.
 
Pebbles : Bidirectional Texture Function
Method : Checkerboard

This video shows the separation results for the pebbles sample as a function of lighting direction (varied from -30 degrees to +30 degrees in steps of 3 degrees, where 0 degrees corresponding to the viewing direction). As the source direction varies, the direct image varies significantly within local regions as the shadows shift and the shading changes. However, the global image varies more smoothly with respect to the lighting direction. MPEG, Quicktime