Sunday, 31 August 2008

Adventures in 3D: Part IV - Let There Be Light

<SwissToni> You see, creating 3D objects is like making love to a beautiful woman</SwissToni>. You've got to set the mood, a bit of gentle lighting here and there, create some contrast.

Simple lighting is actually easy stuff, just wave hello to our friend, the Vector. Remember from Part I that the Vector simply consists of measurements along the x,y,z axes, and you can crunch some numbers to get a couple of useful things out of it, which we'll come to.

Consider a single light source in our scene. It's easy enough to see that any faces that are pointing directly towards the light will be fully lit, and any that are pointing away from the light will be in the dark. In between those, faces that are at some angle towards the light will be partially lit, in some proportion that is a function of that angle.

In 3D space, you can get an idea of how much a given face is pointing in a chosen direction by calculating the dot product. The dot product gives you just a number (a scalar value). If you vectors are of unit length, then the dot product will range from -1 (the vectors are in opposite directions) to 0 (the vectors are at right angles to each other) to +1 (the vectors are in the same direction). The dot product is stupidly easy to work out, it's just multiplication and addition. Given vectors A(x,y,z) and B(x',y',z'), the dot product is defined as:

A.B = x*x' + y*y' + z*z'

Let's start putting this into our code. We'll need to define a Vector class, which is just x,y,z values, plus a method to calculate that dot product.

public class Vector {

public double x;
public double y;
public double z;

public Vector(double x, double y, double z){
this.x = x;
this.y = y;
this.z = z;

public double dotProduct(Vector other){
return this.x * other.x + this.y * other.y + this.z * other.z;

So far, so good. We know we need to calculate the dot product, but what Vectors are we using? One is obviously going to be a Vector representing the direction of the light, and that's simple because we just need to define an arbitrary Vector. What we need to know is how that compares to the direction that each surface in our scene is pointing. For that, we need to calculate the cross product. The cross product of two vectors results in another Vector that points at right angles to those original Vectors. Again, it's startlingly simple to work out. Take A(x,y,z) and B(x',y',z') again, and you can define a new Vector C(x'',y'',z'') where:

x'' = y*z' - z*y'
y'' = z*x' - x*z'
z'' = x*y' - y*x'

So let's add that into our Vector class as well:

public Vector crossProduct(Vector other){

Vector v = new Vector();

v.x = (this.y * other.z) - (this.z * other.y);
v.y = (this.z * other.x) - (this.x * other.z);
v.z = (this.x * other.y) - (this.y * other.x);

return v;

Given two vectors in the plane of our triangle, the cross product will give you a vector that points perpendicular to the triangle (a.k.a the Surface Normal). And how do you get two Vectors in the plane of the triangle? Simple, just work out vectors between the points of the triangle. A Vector is just the difference between two points. We can add a useful method to our Point class to return a Vector given another Point.

public Vector vectorTo(Point p){
return new Vector(this.x - p.x, this.y - p.y, this.z - p.z);

and given that method, we can take the three points in our triangle, calculate two vectors, get their cross product and wa la, you have your surface normal. A bit like this...

 public Vector getNormal(){

Point p1 = new Point(x[0], y[0], z[0]);
Point p2 = new Point(x[1], y[1], z[1]);
Point p3 = new Point(x[2], y[2], z[2]);

Vector edge1 = p2.vectorTo(p1);
Vector edge2 = p3.vectorTo(p2);

return edge1.crossProduct(edge2);

Note that the creation of the Points and use of the vectorTo() method could actually be done away with, and we could just create a vector using e.g. new Vector(x[1]-x[0],y[1]-y[0],z[1]-z[0]), but it helps demonstrate the concepts.

Right, we have our surface normal, and we have our arbitrary light vector. Let's for now say that the light vector L is (1,0,0) i.e. the light is from the right. We're going to get the dot product to see just how parallel they are, and then use that to create a colour to use for the triangle. For now we'll just stick with greyscale, and it makes sense to use Hue,Saturation,Brightness (HSB) instead of RGB - then we can just alter the Brightness based on the dot product. When setting the colour, Brightness needs to be between 0.0 and 1.0, so we'll need to deal with vectors of unit length, otherwise the dot product will be outside that range. That calls for normalisation. To normalise a vector simply means to keep it pointing in the same direction, but make it's length equal to 1. Don't confuse normals (perpendicular vectors) with normalised vectors (vectors scaled to unit length). Let's add a couple of methods to the Vector class.

  public Vector normalise () {
return scale (1.0 / Math.sqrt (dotProduct(this)));

public Vector scale (double scale) {
return new Vector (x * scale, y * scale, z * scale);

The scale() method just scales the vector along each axis. Note the trick to get the length of the vector - the dot product of a vector with itself is the length squared. With these methods, we can now make sure that we're dealing with two unit vectors in the Triangle's draw() method. To summarise the whole thing again - we'll take the triangle's surface normal and normalise it, then take the light vector and normalise it, then find the dot product between the two. Given that they're both normalised, we'll get a number between -1 and 1, which will be used to set the brightness of the colour that's used to draw the polygon.

     double dt = getNormal().normalise().dotProduct(normLight);
Color c = Color.getHSBColor(0.0f, 0.0f, (float) dt);
graphics.fillPolygon(xPoints, yPoints, 3);

FAIL! Well, two thirds fail at least. It's looking fairly cool on the left side, not so hot on the right side. That's because everything on the right hand side is returning a number less than zero, which obviously causes the getHSBColor() method to go a bit crazy (actually, I'm surprised it didn't just bork altogether). So we plainly need to have some sort of check in there to make sure we only deal with positive numbers. Or wait, do we? Remember that we set our light source at (1,0,0), so the light should be coming from the right - right? The thing is, the surface normals for polygons will be pointing to the right. The light vector is coming from the right, pointing left. So actually, the correct polygons to be lit are those that have negative dot products i.e. the vectors are in opposite directions. So let's add a quick check. Anything with a zero or positive dot product is on the dark side of the sphere, so we'll just draw them black.

     Color c = (dt < 0) ?
Color.getHSBColor(0.0f, 0.0f, (float) Math.abs(dt)) :

I like! If you like too, download the source.

Wait a second though. Have a go at rotating the object. Sure, it's prettily lit, but at some point you'll find something a bit odd going on. In Part V - Law And Z-Order, we'll talk Z-order.

No comments: