## July 26, 2011

### Poor Man's Graphics

This is Poor Man's Graphics( or ProgramMers' Graphics). As I observe, some programmers are reluctant to use any graphics program. Whenever they want to explain some vector or geometric stuffs, they would rather do it in ASCII in source code comment.

If you are one of them, I tell you what, Inkscape or Google Draw is not that hard to use. I know I can't convince you, so instead I bring you the concept and the proof of concept of an Ascii to image/ graphics converter - Poor Man's Graphics(PMG).

The best part of it? It is interactive so type in the text boxes below now! And all source code is embedded in this HTML file. Tada- happy coding and happy commenting! remember to toogle 'insert' when necessary.

PMG can only parse the 2 specific types of graphics below. It took me longer than expected, 1 week, to complete the following demo. It is indeed a very good programming exercise. If you have already programmed many apps, games and get bored with programming, then you should try to implement an Ascii art parser like me. It integrates basic concepts of computational pattern matching, geometry and graphics.

Most importantly, there are so many styles you can try!

small challenge

BIG challenge

## July 15, 2011

### Drawing polylines by tessellation

with joints, caps, feathering and varying color

## Introduction

You can consider this as the second episode to the first article Drawing nearly perfect 2D line segments in OpenGL. In a 2D graphics application, drawing only line segments is not enough. We need polylines.

## Analysis Why don't we draw a polyline by a set of line segments?
If we do so, there would be a gap and overdraw at the join between segments. The image on the right are 2 grey segments with 50% transparency. There is a big gap and the darkened part is drawn 2 times. Any polyline thicker than 1.5px will not look good.
To avoid gaps and overdraw(pretend this word as a noun), proper joint treatment is needed. The 3 common joint types as seen in Cairo and most graphics libraries are:
• miter Get the 'outer' border of each line segment and find the intersection point. Use the intersection point as the sharp end. However when the angle between lines is really small, intersection point would be at infinity. As a fall back when the included angle is smaller than a critical value, switch to type bevel.
This not a perfect solution, as, if the polyline is animatable, it would change abruptly from sharp to beveled.
• bevel Take the 'outer' corner of the two lines and connect them to form a bevel.
• round Draw a circle centered at the common vertex with radius half the line width. Sounds easy but if not drawn carefully, there would be serious overdraw.
and common cap types:
• butt The red line is the skeleton of the segment and a butt end always stays inside it.
• round Draw a semi- circle, with radius half the line width, at the end points of a segment.
• square Visually it is the same as a butt end. A segment is extended at a 'square cap', where the extension is equal to half the line width.

## Our approach Any polyline can be broken down into a set of 3-points-polyline ('^' or 'v'), and we call them anchors. In other words, base on an anchor drawing routine, we can build a polyline drawing routine easily.
Note: the approach attempted in this article is not general polygon outward offseting (as in computational geometry) or line buffering (as in GIS). Readers should seek other resources for those approaches.
The work flow to draw a polyline by tessellation:
1. we receive a series of points which make up a polyline, together with color, thickness and additional styles like joint type and cap type
2. break down the polyline into a set of anchors and issue anchor() calls
3. calculate the geometry(outline) of the anchors according to the thickness
4. break down the geometry into triangles with no overlap
5. according to the outline, give each vertex of the triangles a color with alpha
6. output the list of triangles and send them off the rendering pipeline, to be rasterized ultimately.
We will go through these steps one by one now, except 6, to keep this article less OpenGL specific.

## Input

Assume we receives an array of points and color by:
```struct Point
{
double x,y;
//constructors and operator overloadings...
};
struct Color
{
float r,g,b,a;
};
void polyline( const Point* P, const Color* C, int size_of_P,
double weight, char joint_type, char cap_type);```
And the `Point` class has many methods and overloaded operators, allowing we do something like `Point mid_point = (P+P)*0.5;`
We will discover why we are receiving an array of color soon.

## Breaking down a polyline into anchors

There are many possible ways to break, and the simpliest is to break at the mid point of each segment: 1. find the mid point of each segment of the polyline. The mid point between `P` and `P` is referred as mid
2. replace `mid` with `P` and `mid[size_of_P-2]` with `P[size_of_P-1]`
3. for `i=1` to `size_of_P-2`,
create anchor with points`[mid[i-1],P[i],mid[i]]`
The first cap of the first anchor and the last cap of the last anchor must be drawn. No cap for rest of the anchors. That means the anchor() function must allow us to choose which cap to draw.
The declaration of anchor() should then be like:
```void anchor( const Point* P, const Color* C,
double weight, char joint_type, char cap_type,
bool cap_first, bool cap_last);```

## Anchor metrics `P,P,P` the 3 points which make up an anchor `T` the perpendicular vector of the line[`P,P`], pointing from `P` to the outer border of an anchor `T` the perpendicular vector of the line[`P,P`], pointing from `P` to the outer border of an anchor `aT` same as `T` but placed on `P` `bT` same as `T` but placed on `P` `vP` a vector pointing from `P` to the intersection of line [`T',aT'`] and [`T',bT'`], with the 4 vectors placed on their respective points, e.g. `T'=T+P`

To get the outward vector, rotate the vector `T=P-P` anti- clockwise 90 degrees. If the points `P,P,P` are in clockwise order, do nothing. Otherwise, put the vector `T` in the opposite direction. Then normalize `T` and scale to the required thinkness.

Note: all code here uses upper left as origin. Clockwise/ anti- clockwise depends on your chosen coordinate system.
In pseudo- C++ code,
```Point T;
T = P-P;          T = P-P;
T = perpen(T);       T = perpen(T);
if ( signed_area(P,P,P) > 0)
{
T = -T;      T = -T;
}
T = normalize(T);    T = normalize(T);
T *= weight;            T *= weight;

Point perpen(Point P) //perpendicular: anti-clockwise 90 degrees
{
return Point(-P.y,P.x);
}
double signed_area(Point P1, Point P2, Point P3)
{
return (P2.x-P1.x)*(P3.y-P1.y) - (P3.x-P1.x)*(P2.y-P1.y);
}```
To calculate the intersection point between 2 lines, the method is explained here. Say we have an implementation like this: ```int intersect( Point P1, Point P2, //line 1
Point P3, Point P4, //line 2
Point& Pout);       //output point```
Then find vP by:
```Point interP, vP;
intersect( T+P,T+P, T+P,T+P, interP);
vP = interP - P;```
Having all these metrics, we can triangulate anchors for mitered joint and bevelled joint without difficulty, but not a round joint.

## Inner arc

As mentioned before, to avoid overdraw, we cannot simply draw a circle over a round joint. We should only fill the gap by creating an inner arc from aT to bT. An inner arc is the shorter one of the 2 possible arcs between 2 specified angles.
First, lets look at the code for a basic arc:
```void basic_arc( Point P, //origin
float r,      //radius
float dangle, //angle for each step
float angle1, float angle2)
{
bool incremental=true;
if ( angle1>angle2) {
incremental = false; //means decremental
}

if ( incremental) {
for ( float a=angle1; a < angle2; a+=dangle)
{
float x=cos(a);    float y=sin(a);
Point q( P.x+x*r,P.y-y*r); //the current point on the arc
}
} else {
for ( float a=angle1; a > angle2; a-=dangle)
{
float x=cos(a);    float y=sin(a);
Point q( P.x+x*r,P.y-y*r);
}
}
}```
The first trial to fill an arc between 2 vectors:
```void basic_vectors_arc( Point P, //origin
Point A, Point B,
float r) //radius
{
A = normalize(A);          B = normalize(B);
float angle1=acos(A.x);    float angle2=acos(B.x); //A dot x-axis = A.x

basic_arc( P,r,PI/18, angle1,angle2);
}``` It only gives correct result when both A and B are upward. When any one of them is downward, it is wrong, see the interactive demo. One reason is arc cosine returns only from 0 to PI i.e. 0 to 180 degrees. To extend the range to 0 to 2*PI, do this after getting the value of acos():
```if ( A.y>0){ angle1=2*PI-angle1;}
if ( B.y>0){ angle2=2*PI-angle2;}``` An inner arc is always shorter than or equal to a half- circumference. If angle2-angle1 is greater than PI, minus angle2 by 2*PI.
Consider the image on the left. Say angle1=120° and angle2=330°. If the arc is calculated incrementally from angle1 to angle2, it would be an outer arc. Since angle2-angle1=210° > 180°, minus angle2 by 360° and becomes -30°. As defined by basic_arc, the arc is now calculated incrementally from angle2 to angle1, which is an inner arc. Handle similarly when angle1>angle2.
Sample code is at the same place. Then, we can generate a triangle fan between aT and bT for round joint and round cap. The triangulation on the left chose `-vP` as the apparent center of the fan. Anyway, if the color is all the same over an anchor, the form of triangulation does not matter. Otherwise, triangulation does affect color interpolation.
Tips: use `arc length = radius * angle` to control dangle, so that the joint would remain smooth under any thickness, as the number of triangles is made proportional to radius.

## Applying colors

We are receiving an array of color because we want to do per- vertex coloring. There are many profiles of coloring, as much as a child can produce by coloring a car with crayons. Here we just give each vertex the color of its nearest input vertex.   Suggested further work in coloring profile.

## Facing the failed case The above mentioned tessellation method can draw an anchor correctly at most cases. But not when the two segments are making a very small angle, overlapping and start to degenerate into one line segment. At degenerated case, the intersection point `vP` would be at infinity. We now have a slightly differed set of metrics. To identify a degeneration, intersect the green line segment `[T+P, -T+P]` with the red one `[-T+P, -T+P]`. If the intersection point `TP` lies inside both segments, degeneration occurs.
Consider again when the order of points is reversed.
Luckily, the joint is unaffected.

## Fade polygon To achieve anti- aliasing using the 'fade polygon technique' mentioned in the first article, or just to make it more complicated, we can also render the fade polygon of an anchor. The math is the same, so I will not cover it here. An addon is we can arbitrary scale the thickness of the fade polygon to achieve feathering. The effect? Very cool!
image on the right: an implementation of anchor() with round joint, round caps and feathering in OpenGL.

## Introducing Vase Renderer

The implementation of all the above mentioned ideas to render polylines is put into a library called Vase Renderer. It is open sourced. It is still young so the only function it has is polyline().
Vase Renderer is the attempt to create high quality 2D graphics in OpenGL with a different fundament. Instead of thinking about pixels, we think about triangles. It is the attempt to break historical limitations of 2D graphics libraries. For example, Cairo, SVG has no per vertex color. They do not allow variable color along a polyline. It is not they cannot think of this feature (I believe), just it takes so much consideration to support varying color that they would better redesign the library from scratch. 2D computer graphics still needs evolution.  The benefit of tessellating each triangle by hand is you can control the color of each vertex, form of each triangle and overall topology. We can then create gracefull color blending. Moreover, although the implementation process is tough, once it is finished, the result is nice and fast.

## Using the code

For usage and issues about Vase Renderer, visit the current documentation page.

## Limitations

Each anchor is processed separately and is independent of each other. At degeneration overdraw would occur. If the polyline is colored, the artifact is especially obvious.  