Mike Stark

Final Project

Image Synthesis

The final project was to construct a scene of our own choosing, record the scene with the department's digital camera, then render the scene with our ray-tracer making the result as close as possible to the camera image. We were not allowed to reverse engineer the colors, so a significant portion of the project was to "calibrate" the camera, based on the known spectral qualities of the Macbeth Color Checker seen in the images below, illuminated by a light source having a known spectral radiance.

As it happens, the camera calibration is not as easy as one might think. My scene was illuminated with a 75 watt incandescent bulb, which I assumed to be emitting an ideal blackbody (Plank) spectrum at 2800 Kelvins. I have no way of knowing what the emission spectrum really is, but using thar assumptions to approximate the response functions of the camera seems to generate a fairly good approximation to the correct colors. In any case, the image below wouldn't have the correct colors anyway, because it doesn't include any indirect lighting.

Photographed Image

Below is the photographed image. The color checker was sitting on a small table, on a piece of black paper. There is only one direct light source, a 75 watt "soft white" bulb, out of view at the upper left. The sphere is acrylic, with an index of refraction of about 1.47 (this I deduced from using ray-tracing methods.) The glass triangle is a demonstration prism, and the lens is a plano-convex lens made of optical glass. The mirror is something Tushar and I picked up at a craft store.

Rendered Image (Iteration 1)

The image below was rendered with my ray tracer. I used 1.47 for the index of refraction of the sphere (this I came up with by inverse ray-tracing) and 1.52 for the glass in the scene. The large prism has a slight greening, the nature of which I pretty much just guessed at. The lens is made of optical glass so there is no attenuation there. The "caustic" from the sphere, the shadows of the transparent objects, and the general reflection were all computed using "backward" ray tracing, which is from the light to the scene to the camera, rather than from the camera to the scene to the light.

What's Wrong or Missing

The image above is really just a first version. There are a number of things I need to fix, which I hope to do later this summer (after I've taken what I consider a much-needed break!)

There is no "environment mapping", only objects directly visible in the scene are modeled Unfortunately there is so much glass in the scene that certain elements of the environment are visible in reflection. Most notably, the light bulb, part of the lamp arm (and the black paper shielding the rest of the lamp arm and base from view) The light streak near the top of the acrylic sphere is the light track overhead. Lower down on the sphere it looks like the streak continues, but this is actually the reflection of one of the legs of the tripod holding the camera. The tripod is actually more apparent in the reflection of the sphere in the big mirror.
There is no indirect lighting from the environment The backward ray-tracing takes care of the single-bounce indirect lighting from objects in the scene, but the walls of the room are a grayish color, and account for the generally more illuminated glass in the photographed image.
Not all the glass surfaces are polished, although they are modeled as if they are. Most notably, the chamfered edges of the glass prism are "ground" glass. I tried some fine bump-mapping to model this, but it didn't work so well. I think modelling the microgeometry might work (as Pete suggested) but I'll have to understand what the microgeometry is before I do that.
The image was rendered using a "pinhole" camera model. This was actually more of an oversight. I did all the development using a pinhole camera, because it works better with just one sample per pixel, and I didn't remember until just now I forgot to change it. It may take some time to accurately model the camera, however.

Anyway, that's all for later.