Photometric Stereo: PS3 vs PS4

Better results in estimation of surface relief and albedo image

The pair of images shown below are recovered surface relief images obtained using “PS3” and “PS4” for the real texture “rkb1”. Their corresponding recovered surface albedo images are presented too. We may note that the resulting image obtained using “PS4” is better that that obtained using “PS3” and provides more detailed information about either the surface relief or albedo. There are some obvious improvements in the images of “PS4”. For example, looking at the region B1, a better recovered surface relief is apparent in the image for “PS4”, and clarifies the darker region in the image for “PS3” which is due to the effect of shadow. With regard to the albedo image, the region B2 obtained using “PS4” reveals more correct albedo information than that obtained using “PS3”.

Recovered surface relief images obtained from PS3 and PS4

Recovered surface albedo images obtained from PS3 and PS4

Strategy: Photometric stereo using four lights (PS4)

We therefore propose an additional strategy for the photometric stereo technique by using more images (PS4) to estimate surface properties, including both gradient data and albedo data, where we take into account of the effect of shadow. The main advantage of the new method is that surface properties can be estimated more accurately. More images are obtained using further illumination sources, such that more information can be obtained. Our strategy is to treat the shadows as outliers which can be removed. The scheme of this new algorithm of photometric stereo may take the following form:

for each image pixel x (x=0;x<n;x++)
{
select three image pixels xi , xj and xk among four input image pixels i1, i2 , i3 and i4;
{
where xi , xj and xk are the three brightest pixels.
}
do
{
estimating surface derivatives and albedo using selected three pixels of xi , xj and xk from a selective photometric stereo set with three images of Ii , Ij and Ik ;
}
}
end

To achieve this, we

first need to able to distinguish pixel with the most shadow from the pixels in each of the four input images based on a pixel by pixel basis. In this case, the darkest pixel is considered to be a shadowed one and the brightest three pixels are used as a set of photometric stereo data for that specific pixel location. In other words, we assume that the outlier pixel is a shadowed one.
and secondly we have to resolve the surface properties at the corresponding pixel. Shadowed pixels in each set of photometric stereo images can be highlighted in this process with the hope that there are still enough image pixels which are not in shadow. So that we can exclude the shadowed pixel from consideration and recover the surface partial derivatives and albedo using the other three pixels.