Color models

Everybody wants to work with color in their application, but using color raises problems. The most important problem is simply how to represent a color. There are many different ways to encode color information: red, green, blue (RGB) values; hue, saturation, value (HSV); hue, lightness, saturation (HLS); and more. In addition, you can provide full color information for each pixel, or you can just specify an index into a color table (palette) for each pixel. The way you represent a color is called a color model. AWT provides tools for two broad groups of color models: direct and indexed.

As you might expect, you need to specify a color model in order to generate pixel data; the abstract class java.awt.image.ColorModel represents a color model. A ColorModel is one of the arguments to the setPixels() method an image producer calls to deliver pixels to a consumer. What you probably wouldn't expect is that you can use a different color model every time you call setPixels(). Exactly why you'd do this is another matter. Most of the time, you'll want to work with a single color model; that model will probably be the default direct color model. But the additional flexibility is there if you need it.

By default, the core AWT components use a direct color model called ARGB. The A stands for "alpha,", which is the historical name for transparency. RGB refers to the red, green, and blue color components that are combined to produce a single, composite color. In the default ARGB model, each pixel is represented by a 32-bit integer that is interpreted as four 8-bit fields; in order, the fields represent the transparency (A), red, green, and blue components of the color, as shown in Figure 14.3.

To create an instance of the default ARGB model, call the static getRGBdefault() method in ColorModel. This method returns a DirectColorModel object; DirectColorModel is a subclass of ColorModel. You can also create other direct color models by calling a DirectColorModel constructor, but you shouldn't need to unless you have a fairly exotic application.

In an indexed color model, each pixel is represented by a smaller amount of information: an index into a table of real color values. For some applications, generating data with an indexed model may be more convenient. If you have an 8-bit display or smaller, using an indexed model may be more efficient, since your hardware is internally using an indexed color model of some form.

While AWT provides IndexedColorModel objects, we won't cover them in this book. It's sufficient to work with the DirectColorModel. Even if you have an 8-bit display, the Java implementation on your platform should accommodate the hardware you have and, if necessary, dither colors to fit your display. Java also produces transparency on systems that don't natively support it by dithering colors.

Creating an image

Let's take a look at producing some image data. A picture may be worth a thousand words, but fortunately, we can generate a picture in significantly fewer than a thousand words of Java. If we're interested in producing a static image with just one frame, we can use a utility class that acts as an ImageProducer for us.

java.awt.image.MemoryImageSource is a simple utility class that implements the ImageProducer interface; we give it pixel data in an array and it sends that data to an image consumer. A MemoryImageSource can be constructed for a given color model, with various options to specify the type and positioning of its data. We'll use the simplest form, which assumes an ARGB color model.

The following applet, ColorPan, creates an image from an array of integers holding ARGB pixel values:

import java.awt.*;
import java.awt.image.*;
public class ColorPan extends java.applet.Applet { 
    Image img;
    int width, height;
    int [] pixData;
    public void init() {
        width = getSize().width;
        height = getSize().height;
        pixData = new int [width * height];
        int i=0;
        for (int y = 0; y < height; y++) {
            int red = (y * 255) / (height - 1);
            for (int x = 0; x < width; x++) {
                int green = (x * 255) / (width - 1);
                int blue = 128;
                int alpha = 255;
                pixData[i++] = (alpha << 24) | (red << 16) | 
                               (green << 8 ) | blue;
            }
        }
    }
    public void paint( Graphics g ) {
        if ( img == null )
            img = createImage( new MemoryImageSource(width, height, 
                                          pixData, 0, width));
        g.drawImage( img, 0, 0, this );
    }
}

Give it a try. The size of the image is determined by the size of the applet when it starts up. What you should get is a very colorful box that pans from deep blue at the upper left-hand corner to bright yellow at the bottom right with green and red at the other extremes. (You'll have to imagine something more colorful than the black and white image above.)

We create the pixel data for our image in the init()method, and then use MemoryImageSource to create and display the image in paint(). The variable pixData is a one-dimensional array of integers that holds 32-bit ARGB pixel values. In init() we loop over every pixel in the image and assign it an ARGB value. The alpha (transparency) component is always 255, which means the image is opaque. The blue component is always 128, half its maximum intensity. The red component varies from 0 to 255 along the y axis; likewise, the green component varies from 0 to 255 along the x axis. The line below combines these components into an ARGB value:

pixData[i++] = (alpha << 24) | (red << 16) | (green << 8 ) | blue; 

The bitwise left-shift operator (<<) should be familiar to C programmers. It simply shoves the bits over by the specified number of positions. The alpha value takes the top byte of the integer, followed by the red, green, and blue values.

When we construct the MemoryImageSource as a producer for this data, we give it five parameters: the width and height of the image to construct (in pixels), the pixData array, an offset into that array, and the width of each scan line (in pixels). We'll start with the first element (offset 0) of pixData; the width of each scan line and the width of the image are the same. The array pixData has width * height elements, which means it has one element for each pixel.

We create the actual image once, in paint(), using the createImage() method that our applet inherits from Component. In the double-buffering and off-screen drawing examples, we used createImage() to give us an empty off-screen image buffer. Here we use createImage() to generate an image from a specified ImageProducer. createImage() creates the Image object and receives pixel data from the producer to construct the image. Note that there's nothing particularly special about MemoryImageSource; we could use any object that implements the image-producer interface inside of createImage(), including one we wrote ourselves. Once we have the image, we can draw it on the display with the familiar drawImage() method.

You can use MemoryImageSource to produce complex, pixel-by-pixel graphics or render images from arbitrary sources. It produces static images, though; when it reaches the end of its pixel data, its job is done. To generate a stream of image data or update pixel values, we need a more persistent image producer.

Updating a MemoryImageSource

MemoryImageSource can also be used to generate a sequence of images, or to update an image dynamically. In Java 1.1, this is probably the easiest way to build your own low-level animation software. This example simulates the static on a television screen. It generates successive frames of random black and white pixels and displays each frame when it is complete. The figure below shows one frame of random static, followed by the code:

import java.awt.*;
import java.awt.image.*;
public class StaticGenerator 
        extends java.applet.Applet 
        implements Runnable {
    int arrayLength, pixels[];
    MemoryImageSource source;
    Image image;
    int width, height;
    public void init() {
        width = getSize().width; height = getSize().height;
        arrayLength = width * height;
        pixels = new int [arrayLength];
        source = new MemoryImageSource(width, height, pixels, 0, width);
        source.setAnimated(true);
        image = createImage(source);
        new Thread(this).start();
    }
    public void run() {
        while (true) {
            try {
                Thread.sleep(1000/24);
            } catch( InterruptedException e ) { /* die */ }
            for (int x = 0; x < width; x++) 
                for (int y = 0; y < height; y++)  {
                    boolean rand = Math.random() > 0.5;
                    pixels[y*width+x] = 
                        rand ? Color.black.getRGB() : Color.white.getRGB();
                }
            // Push out the new data
            source.newPixels(0, 0, width, height);
        }
    }
    public void paint( Graphics g ) {
        g.drawImage(image, 0, 0, this);
    }
}

The init() method sets up the MemoryImageSource that produces the sequence of images. It first computes the size of the array needed to hold the image data. It then creates a MemoryImageSource object that produces images the width and height of the display, using the default color model (the constructor we use assumes that we want the default). We start taking pixels from the beginning of the pixel array, and scan lines in the array have the same width as the image. Once we have created the MemoryImageSource, we call its setAnimated() method to tell it that we will be generating an image sequence. Then we use the source to create an Image that will display our sequence.

We next start a thread that generates the pixel data. For every element in the array, we get a random number, and set the pixel to black if the random number is greater than 0.5. Because pixels is an int array, we can't assign Color objects to it directly; we use getRGB() to extract the color components from the black and white Color constants. When we have filled the entire array with data, we call the newPixels() method, which delivers the new data to the image.

That's about all there is. We provide a very uninteresting paint() method, that just calls drawImage() to put the current state of the image on the screen. Whenever paint() is called, we see the latest collection of static. The image observer, which is the Applet itself, schedules a call to paint() whenever anything interesting has happened to the image. It's worth noting how simple it is to create this animation. Once we have the MemoryImageSource, we use it to create an image that we treat like any other image. The code that generates the image sequence can be arbitrarily complex--certainly in any reasonable example, it would be more complex than our (admittedly cheesy) static. But that complexity never infects the simple task of getting the image on the screen and updating it.

Image consumers

First, we have to know a little more about the image consumers we'll be feeding. An image consumer implements the seven methods that are defined in the ImageConsumer interface. Two of these methods are overloaded versions of the setPixels() method that accept the actual pixel data for a region of the image. They are identical except that one takes the pixel data as an array of integers, and the other uses an array of bytes. (An array of bytes is natural when you're using an indexed color model because each pixel is specified by an index into a color array.) A call to setPixels() looks something like the following:

setPixels(x, y, width, height, colorModel, pixels, offset, scanLength); 

pixels is the one-dimensional array of bytes or integers that holds the pixel data. Often, you deliver only part of the image with each call to setPixels(). The x, y, width, and height values define the rectangle of the image for which pixels are being delivered. x and y specify the upper left-hand corner of the chunk you're delivering, relative to the upper left-hand corner of the image as a whole. width specifies the width in pixels of the chunk; height specifies the number of scan lines in the chunk. offset specifies the point in pixels at which the data being delivered in this call to setPixels() starts. Finally, scanLength indicates the width of the entire image, which is not necessarily the same as width. The pixels array must be large enough to accommodate width*length+offset elements; if it's larger, any leftover data is ignored.

We haven't said anything yet about the colorModel argument to setPixels(). In our previous example, we drew our image using the default ARGB color model for pixel values; the version of the MemoryImageSource constructor that we used supplied the default color model for us. In this example, we also stick with the default model, but this time we have to specify it explicitly. The remaining five methods of the ImageConsumer interface accept general attributes and framing information about the image:

• setHints()

• setDimensions()

• setProperties()

• setColorModel()

• imageComplete()

Before delivering any data to a consumer, the producer should call the consumer's setHints() method to pass it information about how pixels will be delivered. Hints are specified in the form of flags defined in the ImageConsumer interface. The flags are described in Table 14.2. The consumer uses these hints to optimize the way it builds the image; it's also free to ignore them.

setDimensions() is called to pass the width and height of the image when they are known.

setProperties() is used to pass a hashtable of image properties, stored by name. This method isn't particularly useful without some prior agreement between the producer and consumer about what properties are meaningful. For example, image formats such as GIF and TIFF can include additional information about the image. These image attributes could be delivered to the consumer in the hashtable.

setColorModel() is called to tell the consumer which color model will be used to process most of the pixel data. However, remember that each call to setPixels() also specifies a ColorModel for its group of pixels. The color model specified in setColorModel() is really only a hint that the consumer can use for optimization. You're not required to use this color model to deliver all (or for that matter, any) of the pixels in the image.

The producer calls the consumer's imageComplete() method when it has completely delivered the image or a frame of an image sequence. If the consumer doesn't wish to receive further frames of the image, it should unregister itself from the producer at this point. The producer passes a status flag formed from the flags shown in Table 14.3.

As you can see, the ImageProducer and ImageConsumer interfaces provide a very flexible mechanism for distributing image data. Now let's look at a simple producer.

A sequence of images

The following class, ImageSequence, shows how to implement an ImageProducer that generates a sequence of images. The images are a lot like the ColorPan image we generated a few pages back, except that the blue component of each pixel changes with every frame. This image producer doesn't do anything you couldn't do with a MemoryImageSource. It reads ARGB data from an array, and consults the object that creates the array to give it an opportunity to update the data between each frame.

This is a complex example, so before diving into the code, let's take a broad look at the pieces. The ImageSequence class is an image producer; it generates data and sends it to image consumers to be displayed. To make our design more modular, we define an interface called FrameARGBData that describes how our rendering code provides each frame of ARGB pixel data to our producer. To do the computation and provide the raw bits, we create a class called ColorPanCycle that implements FrameARGBData. This means that ImageSequence doesn't care specifically where the data comes from; if we wanted to draw different images, we could just drop in another class, provided that the new class implements FrameARGBData. Finally, we create an applet called UpdatingImage that includes two image consumers to display the data.

Here's the ImageSequence class:

import java.awt.image.*;
import java.util.*;
public class ImageSequence extends Thread implements ImageProducer {
    int width, height, delay;
    ColorModel model = ColorModel.getRGBdefault();
    FrameARGBData frameData;
    private Vector consumers = new Vector();
    public void run() {
        while ( frameData != null ) {
            frameData.nextFrame();
            sendFrame();
            try {
                sleep( delay );
            } catch ( InterruptedException e ) {}
        }
    }
    public ImageSequence(FrameARGBData src, int maxFPS ) {
        frameData = src;
        width = frameData.size().width;
        height = frameData.size().height;
        delay = 1000/maxFPS;
        setPriority( MIN_PRIORITY + 1 );
    }
    public synchronized void addConsumer(ImageConsumer c) {
        if ( isConsumer( c ) ) 
            return;
        consumers.addElement( c );
        c.setHints(ImageConsumer.TOPDOWNLEFTRIGHT |
                 ImageConsumer.SINGLEPASS );
        c.setDimensions( width, height );
        c.setProperties( new Hashtable() );
        c.setColorModel( model );
    }
    public synchronized boolean isConsumer(ImageConsumer c) {
        return ( consumers.contains( c ) );
    }
    public synchronized void removeConsumer(ImageConsumer c) {
        consumers.removeElement( c );
    }
    public void startProduction(ImageConsumer ic) {
        addConsumer(ic);
    }
    public void requestTopDownLeftRightResend(ImageConsumer ic) { }
    private void sendFrame() {
        for ( Enumeration e = consumers.elements(); e.hasMoreElements();  ) {
            ImageConsumer c = (ImageConsumer)e.nextElement();
            c.setPixels(0, 0, width, height, model, frameData.getPixels(), 0, width);
            c.imageComplete(ImageConsumer.SINGLEFRAMEDONE);
        }
    }
}

The bulk of the code in ImageSequence creates the skeleton we need for implementing the ImageProducer interface. ImageSequence is actually a simple subclass of Thread whose run() method loops, generating and sending a frame of data on each iteration. The ImageSequence constructor takes two items: a FrameARGBData object that updates the array of pixel data for each frame, and an integer that specifies the maximum number of frames per second to generate. We give the thread a low priority (MIN_PRIORITY+1) so that it can't run away with all of our CPU time.

Our FrameARGBData object implements the following interface:

interface FrameARGBData { 
    java.awt.Dimension size(); 
    int [] getPixels(); 
    void nextFrame(); 
} 

In ImageSequence's run() method, we call nextFrame() to compute the array of pixels for each frame. After computing the pixels, we call our own sendFrame() method to deliver the data to the consumers. sendFrame() calls getPixels() to retrieve the updated array of pixel data from the FrameARGBData object. sendFrame() then sends the new data to all of the consumers by invoking each of their setPixels() methods and signaling the end of the frame with imageComplete(). Note that sendFrame() can handle multiple consumers; it iterates through a Vector of image consumers. In a more realistic implementation, we would also check for errors and notify the consumers if any occurred.

The business of managing the Vector of consumers is handled by addConsumer() and the other methods in the ImageProducer interface. addConsumer() adds an item to consumers. A Vector is a perfect tool for this task, since it's an automatically extendable array, with methods for finding out how many elements it has, whether or not a given element is already a member, and so on.

addConsumer() also gives the consumer hints about how the data will be delivered by calling setHints(). This image provider always works from top to bottom and left to right, and makes only one pass through the data. addConsumer() next gives the consumer an empty hashtable of image properties. Finally, it reports that most of the pixels will use the default ARGB color model (we initialized the variable model to ColorModel.getRGBDefault()). In this example, we always start sending image data on the next frame, so startProduction() simply calls addConsumer().

We've discussed the mechanism for communications between the consumer and producer, but I haven't yet told you where the data comes from. We have a FrameARGBData interface that defines how to retrieve the data, but we don't yet have an object that implements the interface. The following class, ColorPanCycle, implements FrameARGBData; we'll use it to generate our pixels:

import java.awt.*;
class ColorPanCycle implements FrameARGBData {
    int frame = 0, width, height;
    private int [] pixels;
    ColorPanCycle ( int w, int h ) {
        width = w;
        height = h;
        pixels = new int [ width * height ];
        nextFrame();
    }
    public synchronized int [] getPixels() {
        return pixels;
    }
    public synchronized void nextFrame() {
        int index = 0;
        for (int y = 0; y < height; y++) {
            for (int x = 0; x < width; x++) {
                int red = (y * 255) / (height - 1);
                int green = (x * 255) / (width - 1);
                int blue = (frame * 10) & 0xff;
                pixels[index++] = 
                    (255 << 24) | (red << 16) | (green << 8) | blue;
            }
        }
        frame++;
    }
    public Dimension size() {
        return new Dimension ( width, height );
    }
}

ColorPanCycle is like our previous ColorPan example, except that it adjusts each pixel's blue component each time nextFrame() is called. This should produce a color cycling effect; as time goes on, the image becomes more blue.

Now let's put the pieces together by writing an applet that displays a sequence of changing images: UpdatingImage. In fact, we'll do better than displaying one sequence. To prove that ImageSequence really can deal with multiple consumers, UpdatingImage creates two components that display different views of the image. Once the mechanism has been set up, it's surprising how little code you need to add additional displays.

import java.awt.*;
import java.awt.image.*;
public class UpdatingImage extends java.applet.Applet { 
    ImageSequence seq;
    public void init() {
        seq = new ImageSequence( new ColorPanCycle(100, 100), 10);
        setLayout( null );
        add( new ImageCanvas( seq, 50, 50 ) );
        add( new ImageCanvas( seq, 100, 100 ) );
        seq.start();
    }
    public void stop() {
        if ( seq != null ) {
            seq.stop();
            seq = null;
        }
    }
}
class ImageCanvas extends Canvas { 
    Image img;
    ImageProducer source;
    ImageCanvas ( ImageProducer p, int w, int h ) {
        source = p;
        setSize( w, h );
    }
    public void update( Graphics g ) {
        paint(g);
    }
    public void paint( Graphics g ) {
        if ( img == null )
            img = createImage( source );
        g.drawImage( img, 0, 0, getSize().width, getSize().height, this );
    }
}

UpdatingImage constructs a new ImageSequence producer with an instance of our ColorPanCycle object as its frame source. It then creates two ImageCanvas components that create and display the two views of our animation. ImageCanvas is a subclass of Canvas; it takes an ImageProducer and a width and height in its constructor and creates and displays an appropriately scaled version of the image in its paint() method. UpdatingImage places the smaller view on top of the larger one for a sort of "picture in picture" effect.

If you've followed the example to this point, you're probably wondering where in the heck is the image consumer. After all, we spent a lot of time writing methods in ImageSequence for the consumer to call. If you look back at the code, you'll see that an ImageSequence object gets passed to the ImageCanvas constructor, and that this object is used as an argument to createImage(). But nobody appears to call addConsumer(). And the image producer calls setPixels() and other consumer methods; but it always digs a consumer out of its Vector of registered consumers, so we never see where these consumers come from.

In UpdatingImage, the image consumer is behind the scenes, hidden deep inside the Canvas--in fact, inside the Canvas' peer. The call to createImage() tells its component (i.e., our canvas) to become an image consumer. Something deep inside the component is calling addConsumer() behind our backs and registering a mysterious consumer, and that consumer is the one the producer uses in calls to setPixels() and other methods. We haven't implemented any ImageConsumer objects in this book because, as you might imagine, most image consumers are implemented in native code, since they need to display things on the screen. There are others though; the java.awt.image.PixelGrabber class is a consumer that returns the pixel data as a byte array. You might use it to save an image. You can make your own consumer do anything you like with pixel data from a producer. But in reality, you rarely need to write an image consumer yourself. Let them stay hidden; take it on faith that they exist.

Now for the next question: How does the screen get updated? Even though we are updating the consumer with new data, the new image will not appear on the display unless the applet repaints it periodically. By now, this part of the machinery should be familiar: what we need is an image observer. Remember that all components are image observers (i.e., the class Component implements ImageObserver). The call to drawImage() specifies our ImageCanvas as its image observer. The default Component class-image-observer functionality then repaints our image whenever new pixel data arrives.

In this example, we haven't bothered to stop and start our applet properly; it continues running and wasting CPU time even when it's invisible. There are two strategies for stopping and restarting our thread. We can destroy the thread and create a new one, which would require recreating our ImageCanvas objects, or we could suspend and resume the active thread. As discussed in Chapter 6, Threads, neither option is particularly difficult.