tags:

views:

48

answers:

1

I'm writing a program which will use scan conversion on triangles to fill in the pixels contained within the triangle.

One thing that has me confused is how to determine the x increment for the right edge of the triangle, or for slopes less than or equal to one.

Here is the code I have to handle left edges with a slope greater than one (obtained from Computer Graphics: Principles and Practice second edition):

for(y=ymin;y<=ymax;y++)
{
    edge.increment+=edge.numerator;
    if(edge.increment>edge.denominator)
    {
        edge.x++;
        edge.increment -= edge.denominator;
    }
}

The numerator is set from (xMax-xMin), and the denominator is set from (yMax-yMin)...which makes sense as it represents the slope of the line. As you move up the scan lines (represented by the y values). X is incremented by 1/(denomniator/numerator) ...which results in x having a whole part and a fractional part.

If the fractional part is greater than one, then the x value has to be incremented by 1 (as shown in edge.increment>edge.denominator).

This works fine for any left handed lines with a slope greater than one, but I'm having trouble generalizing it for any edge, and google-ing has proved fruitless.

Does anyone know the algorithm for that?

+1  A: 

There are 4 cases you need to consider: slope > 1, slope between 0 and 1, slope between -1 and 0, and slope less than -1.

You have coded for slope > 1. If you have slope < -1, you can determine that ahead of time, compute increment to be +1 or -1, and change x++ to x += increment.

For the cases where slope is between -1 and 1, you can have the same loop, only with x and y swapped.

EDIT: It sounds like you're trying to fill a triangle by scan converting two legs simultaneously. That's a special case that you can use, but the general way to scan convert a convex polygon (like a triangle) is to put all the points into a list and then for each scan line, draw a line between the min and max column. Here's some pseudocode for the algorithm:

// create a list of points that is the union of all segments
var points = ScanConvert(x1, y1, x2, y2)
             .Union(ScanConvert(x2, y2, x3, y3))
             .Union(ScanConvert(x1, y1, x3, y3));
// group the points by scan line, finding the leftmost and rightmost point on each line
var scanlines = from p in points
                group p by p.Y into scanline
                select new { y = scanline.Key,
                             x1 = scanline.Min(p => p.X),
                             x2 = scanline.Max(p => p.X) }
// iterate over the scan lines, drawing each segment
foreach (var scanline in scanlines)
    DrawLine(x1, y, x2, y);
Gabe
By swapping x and y, do you mean swapping them in the loop - so you would have (for x=xmin....) and then edge.y? Or do you mean switching them in the calculation of the numerator/denominator values?
Zachary
Zachary: You would have to swap x and y in all locations, including the loop and the numerator/denominator calculation.
Gabe
That doesn't really make a lot of sense to me. What if your left hand edge and your right hand edge are in two different categories of slopes? As far as I can tell you only want to iterate from yMin to yMax one time (filling in pixels as you go), then updating the x value for each edge. If you have to switch x's and y's throughout, that seems like it would be impossible to handle situations where your left edge is something like 2, and your right edge is .5.
Zachary
Zachary: you need to rethink your algorithm to make it more general. See my edit.
Gabe
Thanks for the detailed algorithm, but unfortunately this exercise requires me to use the pixel-by-pixel conversion.
Zachary
I'll go ahead and accept your answer though...as in most normal case it is best.
Zachary