We Keep Coding

Most efficient way to render a dynamic image

I want to show an image (in a JPanel) that must update quickly (about 30 fps). I also want to keep my CPU usage as low as possible.
Each image update will consist of either:
Moving a block of pixels to a new location
Replacing a block of pixels with a new block
The first solution that came to mind was something like this:
private BufferedImage screen;
public void runBlockUpdate(int x, int y, int pieceWidth, int pieceHeight byte[] piece){
ImageIcon imgPiece = new ImageIcon(piece);
Graphics2D g = screen.createGraphics();
g.drawImage(imgPiece.getImage(), x, y, pieceWidth, pieceHeight, null);
repaint();
}
#Override
public void paintComponent(Graphics g) {
g.drawImage(screen, 0, 0, (int) (screenRect.width * screenScale), (int) (screenRect.height * screenScale), this);
}
My main performance concern regards the paint() method. I want to know if there are any more efficient ways of doing this before I fully implement this technique.

You can try to use a swing timer an call the refresh method(JFrame.repaint()) every 30millisec

Java - opaque color

i am trying to draw some lines. Problem is about colors. For example. I have several lines of red color, and than i draw one line of blue color (or reversed). And sometimes, that lines those is more, is opaque for that last one.
I tried to make new color and set color with alpha composite 0.7 - for those more lines, and one color i left default - opaque (alpha 1.0). At first i draw more lines, and than last one. But that lines "overwrite" that one. Is there some solution to fix this problem?
I draw that lines on glasspane.
edit: that code is robust, so it is difficult to post it, and it is one part of thesis.
principle is 2 color for example
Color basicColor;
Color similarColor;
than i have paint method and 2 hashmaps as attributes - some points are stored.
i iterate over this map, remember that one point and similar to him, all other connect with
graphics2D.drawLine(x1,y1,x2,y2) and than change color and paint last one line with another color. I am modifying stroke too, to make it more significant.
I hope it will be enough...
edit2:
i have some Point similarPoint than some robust paint method and here is graphics modifying
iterator iterate over list of points' lists.
Point similar = null;
Iterator<Point> secondIterator;
graphics.setColor(colorOfSimilar);
while (iterator.hasNext()) {
Point point = iterator.next();
if (point.equals(similarPoint)) {
similar = similarPoint;
} else {
secondIterator = secondMap.get(point).iterator();
while (secondIterator.hasNext()) {
Point secondPoint = secondIterator.next();
graphics2D.drawLine(point.getX(), point.getY(),
secondPoint.getX(), secondPoint.getY());
}
}
}
if (similar != null) {
secondIterator = secondMap.get(similar);
graphics2D.setColor(hooverColor);
graphics2D.setStroke(new BasicStroke(2.5f));
while (secondIterator.hasNext()) {
Point secondPoint = secondIterator.next();
graphics2D.drawLine(similar.getX(), similar.getY(),
secondPoint.getX(), secondPoint.getY());
}
graphics2D.setColor(colorOfSimilar);
graphics2D.setStroke(new BasicStroke(1.0f));
}
i wrote it in notepad so sorry about some mistakes (i think brackets etc.), but this is mechanism of modifying, around that is other methods for iterate and other, but it is not important. Problem with stroke doesn´t exist, because at first i did it without stroke.
Thanks for any idea.

The result depends on which compositing rule is specified in the graphics context using setComposite(). This utility may be useful in understanding the various modes. It may also help you in preparing an sscce that exhibits the problem you describe.
Addendum: Here's an example that shows how one might use AlphaComposite.Src mode for this.
import javax.swing.*;
import java.awt.*;
import java.awt.geom.*;
/** #see http://stackoverflow.com/questions/7823631 */
public class X extends JPanel {
private static final int SIZE = 300;
private static final int INSET = 64;
private static final AlphaComposite OVER_HALF =
AlphaComposite.getInstance(AlphaComposite.SRC_OVER, 0.5f);
private boolean src;
public X(boolean src) {
this.src = src;
this.setBackground(Color.lightGray);
}
#Override
public Dimension getPreferredSize() {
return new Dimension(SIZE, SIZE);
}
#Override
public void paintComponent(Graphics g) {
super.paintComponent(g);
Graphics2D g2 = (Graphics2D) g;
g2.setRenderingHint(
RenderingHints.KEY_ANTIALIASING,
RenderingHints.VALUE_ANTIALIAS_ON);
Line2D line1 = new Line2D.Double(INSET, INSET,
getWidth() - INSET, getHeight() - INSET);
Line2D line2 = new Line2D.Double(getWidth() - INSET,
INSET, INSET, getHeight() - INSET);
g2.setStroke(new BasicStroke(64,
BasicStroke.CAP_ROUND,
BasicStroke.JOIN_BEVEL));
g2.setComposite(OVER_HALF);
g2.setColor(Color.red);
g2.draw(line1);
if (src) {
g2.setComposite(AlphaComposite.Src);
}
g2.setColor(Color.blue);
g2.draw(line2);
}
public static void main(String[] args) {
final JFrame frame = new JFrame();
frame.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
frame.setLayout(new GridLayout(1, 0));
frame.add(new X(false));
frame.add(new X(true));
frame.pack();
frame.setLocationRelativeTo(null);
frame.setVisible(true);
}
}

java applet: is there a simple way to draw and erase in a two-layered 2D scene?

I have an applet which displays some data using circles and lines. As the data continually changes, the display is updated, which means that sometimes the circles and lines must be erased, so I just draw them in white (my background color) to erase them. (There are a lot of them, so erasing everything and then recomputing and redrawing everything except the erased item would be a horribly slow way to erase a single item.)
The logic of the situation is that there are two layers that need to be displayed, and I need to be able to erase an object in one layer without affecting the other layer. I suppose the upper layer would need to have a background color of "transparent", but then how would I erase an object, since drawing in a transparent color has no effect.
What distinguishes this situation from all the transparency-related help on the web is that I want to be able to erase lines and circles one-by-one from the transparent layer, overwriting their pixels with the "fully transparent" color.
Currently my applet draws (using just a single layer) by doing this in start():
screenBuffer = createImage(640, 480);
screenBufferGraphics = screenBuffer.getGraphics();
and this in paint():
g.drawImage(screenBuffer, 0, 0, this);
and objects are rendered (or "erased" by drawing in white) by commands like:
screenBufferGraphics.drawLine(x1,y1,x2,y2);
Is it easy to somehow make a second screen buffer with a transparent background and then be able to draw and erase objects in that buffer and render it over the first buffer?

This seems fairly quick, so long as the rendered image area remains around 640x480, the code can achieve from 125-165 FPS. The code tracks 2000 semi-transparent lines of width 4px, and moves them around in an area 8 times the size of the rendered image.
import java.awt.image.BufferedImage;
import java.awt.event.*;
import java.awt.geom.*;
import java.awt.*;
import javax.swing.*;
import java.util.Random;
class LineAnimator {
public static void main(String[] args) {
final int w = 640;
final int h = 480;
final RenderingHints hints = new RenderingHints(
RenderingHints.KEY_ANTIALIASING,
RenderingHints.VALUE_ANTIALIAS_ON
);
hints.put(
RenderingHints.KEY_ALPHA_INTERPOLATION,
RenderingHints.VALUE_ALPHA_INTERPOLATION_QUALITY
);
final BufferedImage bi = new BufferedImage(w,h, BufferedImage.TYPE_INT_RGB);
final JLabel l = new JLabel(new ImageIcon(bi));
final BouncingLine[] lines = new BouncingLine[20000];
for (int ii=0; ii<lines.length; ii++) {
lines[ii] = new BouncingLine(w*8,h*8);
}
final Font font = new Font("Arial", Font.BOLD, 30);
ActionListener al = new ActionListener() {
int count = 0;
long lastTime;
String fps = "";
public void actionPerformed(ActionEvent ae) {
count++;
Graphics2D g = bi.createGraphics();
g.setRenderingHints(hints);
g.clearRect(0,0,w,h);
for (int ii=0; ii<lines.length; ii++) {
lines[ii].move();
lines[ii].paint(g);
}
if ( System.currentTimeMillis()-lastTime>1000 ) {
lastTime = System.currentTimeMillis();
fps = count + " FPS";
count = 0;
}
g.setColor(Color.YELLOW);
g.setFont(font);
g.drawString(fps,10,30);
l.repaint();
g.dispose();
}
};
Timer timer = new Timer(1,al);
timer.start();
JOptionPane.showMessageDialog(null, l);
System.exit(0);
}
}
class BouncingLine {
private final Color color;
private static final BasicStroke stroke = new BasicStroke(4);
private static final Random random = new Random();
Line2D line;
int w;
int h;
int x1;
int y1;
int x2;
int y2;
BouncingLine(int w, int h) {
line = new Line2D.Double(random.nextInt(w),random.nextInt(h),random.nextInt(w),random.nextInt(h));
this.w = w;
this.h = h;
this.color = new Color(
128+random.nextInt(127),
128+random.nextInt(127),
128+random.nextInt(127),
85
);
x1 = (random.nextBoolean() ? 1 : -1);
y1 = (random.nextBoolean() ? 1 : -1);
x2 = -x1;
y2 = -y1;
}
public void move() {
int tx1 = 0;
if (line.getX1()+x1>0 && line.getX1()+x1<w) {
tx1 = (int)line.getX1()+x1;
} else {
x1 = -x1;
tx1 = (int)line.getX1()+x1;
}
int ty1 = 0;
if (line.getY1()+y1>0 && line.getY1()+y1<h) {
ty1 = (int)line.getY1()+y1;
} else {
y1 = -y1;
ty1 = (int)line.getY1()+y1;
}
int tx2 = 0;
if (line.getX2()+x2>0 && line.getX2()+x2<w) {
tx2 = (int)line.getX2()+x2;
} else {
x2 = -x2;
tx2 = (int)line.getX2()+x2;
}
int ty2 = 0;
if (line.getY2()+y2>0 && line.getY2()+y2<h) {
ty2 = (int)line.getY2()+y2;
} else {
y2 = -y2;
ty2 = (int)line.getY2()+y2;
}
line.setLine(tx1,ty1,tx2,ty2);
}
public void paint(Graphics g) {
Graphics2D g2 = (Graphics2D)g;
g2.setColor(color);
g2.setStroke(stroke);
//line.set
g2.draw(line);
}
}
Update 1
When I posted that code, I thought you said 100s to 1000s, rather than 1000s to 100,000s! At 20,000 lines the rate drops to around 16-18 FPS.
Update 2
..is this optimized approach, using layers, possible in Java?
Sure. I use that technique in DukeBox - which shows a funky plot of the sound it is playing. It keeps a number of buffered images.
Background. A solid color in a non-transparent image.
Old Traces. The older sound traces as stretched or faded from the original positions. Has transparency, to allow the BG to show.
Latest Trace. Drawn on top of the other two. Has transparency.

After a day of no proposed solutions, I started to think that Java Graphics cannot erase individual items back to a transparent color. But it turns out that the improved Graphics2D, together with BufferedImage and AlphaComposite, provide pretty much exactly the functionality I was looking for, allowing me to both draw shapes and erase shapes (back to full transparency) in various layers.
Now I do the following in start():
screenBuffer = new BufferedImage(640, 480, BufferedImage.TYPE_INT_ARGB);
screenBufferGraphics = screenBuffer.createGraphics();
overlayBuffer = new BufferedImage(640, 480, BufferedImage.TYPE_INT_ARGB);
overlayBufferGraphics = overlayBuffer.createGraphics();
I have to use new BufferedImage() instead of createImage() because I need to ask for alpha. (Even for screenBuffer, although it is the background -- go figure!) I use createGraphics() instead of getGraphics() just because my variable screenBufferGraphics is now a Graphics2D object instead of just a Graphics object. (Although casting back and forth works fine too.)
The code in paint() is barely different:
g.drawImage(screenBuffer, 0, 0, null);
g.drawImage(overlayBuffer, 0, 0, null);
And objects are rendered (or erased) like this:
// render to background
screenBufferGraphics.setColor(Color.red);
screenBufferGraphics.fillOval(80,80, 40,40);
// render to overlay
overlayBufferGraphics.setComposite(AlphaComposite.SrcOver);
overlayBufferGraphics.setColor(Color.green);
overlayBufferGraphics.fillOval(90,70, 20,60);
// render invisibility onto overlay
overlayBufferGraphics.setComposite(AlphaComposite.DstOut);
overlayBufferGraphics.setColor(Color.blue);
overlayBufferGraphics.fillOval(70,90, 30,20);
// and flush just this locally changed region
repaint(60,60, 80,80);
The final Color.blue yields transparency, not blueness -- it can be any color that has no transparency.
As a final note, if you are rendering in a different thread from the AWT-EventQueue thread (which you probably are if you spend a lot of time rendering but also need to have a responsive interface), then you will want to synchronize the above code in paint() with your rendering routine; otherwise the display can wind up in a half-drawn state.
If you are rendering in more than one thread, you will need to synchronize the rendering routine anyway so that the Graphics2D state changes do not interfere with each other. (Or maybe each thread could have its own Graphics2D object drawing onto the same BufferedImage -- I didn't try that.)
It looks so simple, it's hard to believe how long it took to figure out how to do this!

Java: how to do double-buffering in Swing?

EDIT TWO
To prevent snarky comments and one-line answers missing the point: IFF it is as simple as calling setDoubleBuffered(true), then how do I get access to the current offline buffer so that I can start messing with the BufferedImage's underlying pixel databuffer?
I took the time to write a running piece of code (which looks kinda fun too) so I'd really appreciate answers actually answering (what a shock ;) my question and explaining what/how this is working instead of one-liners and snarky comments ;)
Here's a working piece of code that bounces a square across a JFrame. I'd like to know about the various ways that can be used to transform this piece of code so that it uses double-buffering.
Note that the way I clear the screen and redraw the square ain't the most efficient but this is really not what this question is about (in a way, it's better for the sake of this example that it is somewhat slow).
Basically, I need to constantly modify a lot pixels in a BufferedImage (as to have some kind of animation) and I don't want to see the visual artifacts due to single-buffering on screen.
I've got a JLabel whose Icon is an ImageIcon wrapping a BufferedImage. I want to modify that BufferedImage.
What has to be done so that this becomes double-buffered?
I understand that somehow "image 1" will be shown while I'll be drawing on "image 2". But then once I'm done drawing on "image 2", how do I "quickly" replace "image 1" by "image 2"?
Is this something I should be doing manually, like, say, by swapping the JLabel's ImageIcon myself?
Should I be always drawing in the same BufferedImage then do a fast 'blit' of that BufferedImage's pixels in the JLabel's ImageIcon's BufferedImage? (I guess no and I don't see how I could "synch" this with the monitor's "vertical blank line" [or equivalent in flat-screen: I mean, to 'synch' without interfering with the moment the monitor itselfs refreshes its pixels, as to prevent shearing]).
What about the "repaint" orders? Am I suppose to trigger these myself? Which/when exactly should I call repaint() or something else?
The most important requirement is that I should be modifying pixels directly in the images's pixel databuffer.
import javax.swing.*;
import java.awt.event.WindowAdapter;
import java.awt.event.WindowEvent;
import java.awt.image.BufferedImage;
import java.awt.image.DataBufferInt;
public class DemosDoubleBuffering extends JFrame {
private static final int WIDTH = 600;
private static final int HEIGHT = 400;
int xs = 3;
int ys = xs;
int x = 0;
int y = 0;
final int r = 80;
final BufferedImage bi1;
public static void main( final String[] args ) {
final DemosDoubleBuffering frame = new DemosDoubleBuffering();
frame.addWindowListener(new WindowAdapter() {
public void windowClosing( WindowEvent e) {
System.exit(0);
}
});
frame.setSize( WIDTH, HEIGHT );
frame.pack();
frame.setVisible( true );
}
public DemosDoubleBuffering() {
super( "Trying to do double buffering" );
final JLabel jl = new JLabel();
bi1 = new BufferedImage( WIDTH, HEIGHT, BufferedImage.TYPE_INT_ARGB );
final Thread t = new Thread( new Runnable() {
public void run() {
while ( true ) {
move();
drawSquare( bi1 );
jl.repaint();
try {Thread.sleep(10);} catch (InterruptedException e) {}
}
}
});
t.start();
jl.setIcon( new ImageIcon( bi1 ) );
getContentPane().add( jl );
}
private void drawSquare( final BufferedImage bi ) {
final int[] buf = ((DataBufferInt) bi.getRaster().getDataBuffer()).getData();
for (int i = 0; i < buf.length; i++) {
buf[i] = 0xFFFFFFFF; // clearing all white
}
for (int xx = 0; xx < r; xx++) {
for (int yy = 0; yy < r; yy++) {
buf[WIDTH*(yy+y)+xx+x] = 0xFF000000;
}
}
}
private void move() {
if ( !(x + xs >= 0 && x + xs + r < bi1.getWidth()) ) {
xs = -xs;
}
if ( !(y + ys >= 0 && y + ys + r < bi1.getHeight()) ) {
ys = -ys;
}
x += xs;
y += ys;
}
}
EDIT
This is not for a full-screen Java application, but a regular Java application, running in its own (somewhat small) window.

---- Edited to address per pixel setting ----
The item blow addresses double buffering, but there's also an issue on how to get pixels into a BufferedImage.
If you call
WriteableRaster raster = bi.getRaster()
on the BufferedImage it will return a WriteableRaster. From there you can use
int[] pixels = new int[WIDTH*HEIGHT];
// code to set array elements here
raster.setPixel(0, 0, pixels);
Note that you would probably want to optimize the code to not actually create a new array for each rendering. In addition, you would probably want to optimized the array clearing code to not use a for loop.
Arrays.fill(pixels, 0xFFFFFFFF);
would probably outperform your loop setting the background to white.
---- Edited after response ----
The key is in your original setup of the JFrame and inside the run rendering loop.
First you need to tell SWING to stop Rasterizing whenever it wants to; because, you'll be telling it when you're done drawing to the buffered image you want to swap out in full. Do this with JFrame's
setIgnoreRepaint(true);
Then you'll want to create a buffer strategy. Basically it specifies how many buffers you want to use
createBufferStrategy(2);
Now that you tried to create the buffer strategy, you need to grab the BufferStrategy object as you will need it later to switch buffers.
final BufferStrategy bufferStrategy = getBufferStrategy();
Inside your Thread modify the run() loop to contain:
...
move();
drawSqure(bi1);
Graphics g = bufferStrategy.getDrawGraphics();
g.drawImage(bi1, 0, 0, null);
g.dispose();
bufferStrategy.show();
...
The graphics grabbed from the bufferStrategy will be the off-screen Graphics object, when creating triple buffering, it will be the "next" off-screen Graphics object in a round-robin fashion.
The image and the Graphics context are not related in a containment scenario, and you told Swing you'd do the drawing yourself, so you have to draw the image manually. This is not always a bad thing, as you can specify the buffer flipping when the image is fully drawn (and not before).
Disposing of the graphics object is just a good idea as it helps in garbage collection. Showing the bufferStrategy will flip buffers.
While there might have been a misstep somewhere in the above code, this should get you 90% of the way there. Good luck!
---- Original post follows ----
It might seem silly to refer such a question to a javase tutorial, but have you looked into BufferStrategy and BufferCapatbilites?
The main issue I think you are encountering is that you are fooled by the name of the Image. A BufferedImage has nothing to do with double buffering, it has to do with "buffering the data (typically from disk) in memory." As such, you will need two BufferedImages if you wish to have a "double buffered image"; as it is unwise to alter pixels in image which is being shown (it might cause repainting issues).
In your rendering code, you grab the graphics object. If you set up double buffering according to the tutorial above, this means you will grab (by default) the off-screen Graphics object, and all drawing will be off-screen. Then you draw your image (the right one of course) to the off-screen object. Finally, you tell the strategy to show() the buffer, and it will do the replacement of the Graphics context for you.

Generally we use Canvas class which is suitable for animation in Java. Anyhoo, following is how you achieve double buffering:
class CustomCanvas extends Canvas {
private Image dbImage;
private Graphics dbg;
int x_pos, y_pos;
public CustomCanvas () {
}
public void update (Graphics g) {
// initialize buffer
if (dbImage == null) {
dbImage = createImage (this.getSize().width, this.getSize().height);
dbg = dbImage.getGraphics ();
}
// clear screen in background
dbg.setColor (getBackground ());
dbg.fillRect (0, 0, this.getSize().width, this.getSize().height);
// draw elements in background
dbg.setColor (getForeground());
paint (dbg);
// draw image on the screen
g.drawImage (dbImage, 0, 0, this);
}
public void paint (Graphics g)
{
g.setColor (Color.red);
g.fillOval (x_pos - radius, y_pos - radius, 2 * radius, 2 * radius);
}
}
Now you can update the x_pos and y_pos from a thread, followed by the 'repaint' call on the canvas object. The same technique should work on a JPanel as well.

What you want is basically impossible in windowed mode with Swing. There is no support for raster synchronization for window repaints, this is only available in fullscreen mode (and even then may not be supported by all platforms).
Swing components are double-buffered by default, that is they will do all the rendering to an intermediate buffer and that buffer is then finally copied to the screen, avoiding flicker from background clearing and then painting on top of it.
And thats the only strategy that is reasonable well supported on all underlying platforms. It avoids only repaint flickering, but not visual tearing from moving graphic elements.
A reasonably simple way of having access to the raw pixels of an area fully under you control would be to extend a custom component from JComponent and overwrite its paintComponent()-method to paint the area from a BufferedImage (from memory):
public class PixelBufferComponent extends JComponent {
private BufferedImage bufferImage;
public PixelBufferComponent(int width, int height) {
bufferImage = new BufferedImage(width, height, BufferedImage.TYPE_INT_RGB);
setPreferredSize(new Dimension(width, height));
}
public void paintComponent(Graphics g) {
g.drawImage(bufferImage, 0, 0, null);
}
}
You can then manipulate you buffered image whichever way you desire. To get your changes made visible on screen, simply call repaint() on it. If you do the pixel manipulation from a thread other than the EDT, you need TWO buffered images to cope with race conditions between the actual repaint and your manipulation thread.
Note that this skeleton will not paint the entire area of the component when used with a layout manager that stretches the component beyond its preferred size.
Note also, the buffered image approach mostly only makes sense if you do real low level pixel manipulation via setRGB(...) on the image or if you directly access the underlying DataBuffer directly. If you can do all the manipulations using Graphics2D's methods, you could do all the stuff in the paintComponent method using the provided graphics (which is actually a Graphics2D and can be simply casted).

Here's a variation in which all drawing takes place on the event dispatch thread.
Addendum:
Basically, I need to constantly modify a lot pixels in a BufferedImage…
This kinetic model illustrates several approaches to pixel animation.
import java.awt.Color;
import java.awt.Dimension;
import java.awt.EventQueue;
import java.awt.Graphics2D;
import java.awt.GridLayout;
import java.awt.event.ActionEvent;
import java.awt.event.ActionListener;
import javax.swing.*;
import java.awt.image.BufferedImage;
/** #see http://stackoverflow.com/questions/4430356 */
public class DemosDoubleBuffering extends JPanel implements ActionListener {
private static final int W = 600;
private static final int H = 400;
private static final int r = 80;
private int xs = 3;
private int ys = xs;
private int x = 0;
private int y = 0;
private final BufferedImage bi;
private final JLabel jl = new JLabel();
private final Timer t = new Timer(10, this);
public static void main(final String[] args) {
EventQueue.invokeLater(new Runnable() {
#Override
public void run() {
JFrame frame = new JFrame();
frame.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
frame.add(new DemosDoubleBuffering());
frame.pack();
frame.setVisible(true);
}
});
}
public DemosDoubleBuffering() {
super(true);
this.setLayout(new GridLayout());
this.setPreferredSize(new Dimension(W, H));
bi = new BufferedImage(W, H, BufferedImage.TYPE_INT_ARGB);
jl.setIcon(new ImageIcon(bi));
this.add(jl);
t.start();
}
#Override
public void actionPerformed(ActionEvent e) {
move();
drawSquare(bi);
jl.repaint();
}
private void drawSquare(final BufferedImage bi) {
Graphics2D g = bi.createGraphics();
g.setColor(Color.white);
g.fillRect(0, 0, W, H);
g.setColor(Color.blue);
g.fillRect(x, y, r, r);
g.dispose();
}
private void move() {
if (!(x + xs >= 0 && x + xs + r < bi.getWidth())) {
xs = -xs;
}
if (!(y + ys >= 0 && y + ys + r < bi.getHeight())) {
ys = -ys;
}
x += xs;
y += ys;
}
}

How can I repaint efficiently when using big custom component in Swing?

I have made a custom component (derived from JComponent) which represents
a draggable Bezier-curve.
(looks like a hanging cable, someone might know it
from Bender or Cubase)
My problem is: The curve may become really long,
let's say from top left to bottom right corners of the desktop.
This makes Swing's repaint functionality inefficient:
The area of the curve is perhaps few hundred pixels, but the area of
the component (being mostly 'transparent') is millions of pixels big.
My subjection impression is:
The longer the curve, the more flicker I get when dragging it.
I hope I made myself clear about the problem.
Perhaps it would help when I somehow could choose by myself, which regions
of the component needs repainting at all.
EDIT:
Such a mess! I'm profiling the application using Netbeans, which helps to
find inefficient code normally, but this Swing framework is making hundreds
of nested calls! I just can't figure out, what is slow and why.
By the way, disabling super.paint(...) or super.paintComponent(...) doesn't help.

Check out Filthy Rich Clients by Chet Haase and Romain Guy. They address these very optimizations among others along the way to producing responsive and graphically impressive UI.

Doing all of your bezier mathematics on the paint thread everytime the component is refreshed is (as you've gathered) a bad idea. Does your curve change often? If not then why not paint it to a BufferedImage as and when it changes, and change your paint() code to simply draw the buffered image to the component instead.
class CurveComponent extends JComponent {
private BufferedImage image;
#Override
public void paintComponent( Graphics g ) {
if ( image == null ) {
return;
}
g.drawImage( image, 0, 0, this );
}
private void updateCurve() {
image = new BufferedImage( getWidth(), getHeight(), BufferedImage.ARGB );
Graphics g = image.getGraphics();
// draw the curve onto image using g.
g.dispose();
}
}
Only call updateCurve() when you need to and all that expensive mathematics won't be needlessly repeated. Painting should be pretty responsive, even for a fullscreen window. drawImage() will be doing a straightforward memory copy and should be lightning fast.

Try writing a tiny test app, which consists of nothing except what you need to reproduce this problem. This will make profiling easier. Then post that app here, so we can take a look at possible solutions.
I found your question interesting so I wrote a test app myself. This draws a Bezier curve which is continually resized as you drag. I created a gradient background to ensure this works well with a nasty background. I get good performance and low flicker, although I use top-notch machine.
It pays to read "Filthy Rich Clients" to learn all the tricks of writing custom Swing components that perform really well.
import javax.swing.*;
import java.awt.*;
import java.awt.event.MouseAdapter;
import java.awt.event.MouseEvent;
import java.awt.geom.CubicCurve2D;
import java.awt.geom.Point2D;
public class CustomComponent extends JComponent {
private Point2D start = new Point2D.Double(0, 0);
private Point2D end = new Point2D.Double(300, 200);
private CustomComponent() {
this.setOpaque(true);
final MouseAdapter mouseAdapter = new MouseAdapter() {
#Override
public void mouseDragged(MouseEvent e) {
setEnd(e.getPoint());
}
};
this.addMouseListener(mouseAdapter);
this.addMouseMotionListener(mouseAdapter);
}
public void setStart(Point2D start) {
this.start = start;
repaint();
}
public void setEnd(Point2D end) {
this.end = end;
repaint();
}
#Override
protected void paintComponent(Graphics g) {
final Graphics2D g2 = (Graphics2D) g;
g2.setRenderingHint(RenderingHints.KEY_ANTIALIASING, RenderingHints.VALUE_ANTIALIAS_ON);
// draw gradient background
final int width = getWidth();
final int height = getHeight();
g2.setPaint(new GradientPaint(0, 0, Color.WHITE, width, height, Color.YELLOW));
g2.fillRect(0, 0, width, height);
// draw Bezier curve
final Shape shape = new CubicCurve2D.Double(start.getX(), start.getY(), start.getX(), end.getY(), end.getX(), start.getY(), end.getX(), end.getY());
g2.setColor(Color.BLACK);
g2.draw(shape);
g2.drawString("Click and drag to test for flickering", 100, 20);
}
public static void main(String[] args) {
final CustomComponent component = new CustomComponent();
final Dimension screenSize = Toolkit.getDefaultToolkit().getScreenSize();
final Dimension size = new Dimension(screenSize.width - 20, screenSize.height - 100);
component.setPreferredSize(size);
final JFrame frame = new JFrame();
frame.add(component);
frame.pack();
frame.setLocationRelativeTo(null);
frame.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
frame.setVisible(true);
}
}
Some things to note:
only overwrite paintComponent(Graphics g), not the other paintXXX() methods
set custom component to opaque if possible
only use repaint() to request repainting. Never directly order a repaint directly in your code. This lets Swing handle it well.

There is no efficient way to create lots of small clip rectangles for a diagonal structure which leaves you with two strategies to avoid flickering:
Double buffering. This needs an enormous amount of memory but the memory copy is very fast (it usually happens in the time the "electron beam" goes back from lower right to upper left ... if there was still a beam in your LCD).
Don't call super.paint() (which draws or "erases" the background) and draw the curve a second time with the background color to erase it.
For more details, see this document.
[EDIT] If fillRect() wasn't abstract, you could set a break point :) Set a break point in paint(), check who calls it and whether the background got cleared at that time. It should be since rendering would be completely wrong. Then set break points further up in the call chain.

You can redraw a smaller portion of the screen using repaint(Rectangle r)
http://java.sun.com/j2se/1.4.2/docs/api/javax/swing/JComponent.html#repaint(java.awt.Rectangle)
Then you mention flicker. Since you are using swing, which uses double buffering your flickering must be coming from something else. Are you clearing the screen in paintComponent(...)? I.e. call to fillRect(...)? Don't do that, it's not needed (IIRC).

Which method do yo use to paint your curve? paint or paintComponent?

My solution was a partial re-design:
Now I don't represent each "cable"-element by a component.
Now, cables are just dummy objects (with no involved JComponent).
The repaint takes place "globally", on the content pane of the parent JFrame.
Now it's efficient, and flickers less.

just use getVisibleRect(); inside paintComponent(Graphics g) to get the area you actually need to redraw