|
|
|
|
|
|
| Original Image | Diffuse Component | Specular Component |
|
|
|
|
|
|
|
|
| Original Image | Diffuse Component | Specular Component |
We begin by showing results on 12-bit images acquired under laboratory conditions where the light source color was known. The fine texture under the specularity is revealed after the separation.
|
|
|
| Original Image | Diffuse Component | Specular Component |
The diffuse and specular components of a human face are separated. The 12-bit image was acquired outside laboratory conditions and the light source color was not known.
|
|
|
|
|
|
|
| Original Image | Diffuse Component | Specular Component |
The above panels show the result on an 8-bit JPEG image
acquired from the internet. This demonstrates the practical utility of
the algorithm for real world examples.
|
|
|
Move mouse over image |
The animation above shows a human torso. This animation illustrates how the presence of specularity is not obvious until it is removed. See for example, the specularity on the arm. ( Move mouse over the image to see the effect. ) Click here to see a video (DivX encoded AVI, 344KB ) showing the iterative erosion of specularity.
 |
|
Click on the image to download the DivX encoded AVI file. (1.8 MB) |
The above panel shows results of specularity removal in videos. ( Click on the image to download the associated video. ) The technique for video extends the technique for images by using the spatio-temporal information in a video instead of processing each frame independently. To the best of our knowledge, specularity removal on videos has not been demonstrated prior to this work. The top row shows a few original frames. The bottom row shows the diffuse component.
|
||
|
Original Image |
Ground Truth | Our Diffuse Component |
The above panel shows a comparison of our method to ground truth. The original image is shown on the left while the diffuse component obtained using our algorithm is shown on the right. The ground truth (middle image) was obtained using a horizontal polarizer on the light source and a vertical polarizer on the camera.
The diffuse and specular components of an image have very different reflectance properties and form two distinct factors in perception. Therefore, in many situations these two components of a surface are tweaked independently to achieve a desired appearance. For example, many paints are made by mixing two components; one of them determines the color and the other determines the gloss. (e.g. flat, eggshell, satin, semi-gloss and hi-gloss). Similarly, the purpose of makeup on human face is to modify one or both of the reflection components. Matte makeup on a human face aims to smooth the diffuse component and diminish the specular component. On the other hand, a moist appearance on the face is primarily a change in its specular component. With the ability to separate the two components we can independently process and recombine them to simulate many of these different visual effects. We call this layer based processing Dichromatic Editing.
|
The panel shows a few dichromatic editing schemes. In [A - D] the specular component of the image is gradually decreased from twice the specularity in the original image to no specularity. [E - H] shows a scheme to change the skin color. In E the skin color is made fairer by decreasing the saturation of the diffuse component. In F) the specular component is added back to give the image a more natural look. Similarly a darkening of the skin is achieved by changing the value (i.e. the third HSV channel) of the diffuse component in the HSV space. The specularity is added back in H. In I, only the diffuse color of the image is sharpened while in J only the specular component is sharpened. In K Gaussian filtering is applied only to the specular component which gives the effect of soft lighting. In L the diffuse component of the image is smoothed using a Gaussian filter giving an appearance of makeup.
 |
|
 |
 |
|
[A] Original Image |
[B] Diffuse Component |
[C] Specular Component |
[D] Eroding Specular Layer |
 |
 |
 |
|
|
[E] Dilating Specular Layer |
[F] Reflecting Specularity |
[G] Modulating Specularity |
[H] Masking Specularity |
The above panel shows a pear in which the specular component is edited to achieve different visual perceptions. [A] shows the original image, [B] the diffuse component and [C] the specular component. In [D] the specular component was eroded using a disk shaped structuring element and then amplified. The specular BRDF of the pear appears to have a narrow lobe. Similarly in [E] dilation of the specular component results in a broader specular lobe. Noting approximate symmetry of the pear, the specular component was reflected above the vertical axis and added to the original image. This simulates a second light in the scene in [F]. In [G] the pear is given an avocado like appearance by modulating the specular component. In [H] the specular layer is masked giving an appearance of a plastic ribbon tied on a diffuse pear.
