Java RenderedImage实现与自定义原始DataBuffer

Question

I receive 16-bit grayscale images from a device, the images are delivered in an uncompressed raw format , here is a 8 bytes example of how 2X2 image will look like using this format (MSB first) :

21 27           33 F6          28 F3           27 F2
-----           -----          -----           -----
pixel 0,0(x,y)  pixel 1,0      pixel 1,0       pixel 1,1

I need to compress the images using Kakadu JPEG2000 library that expose a Java ImageWriter implementation, the ImageWriter.write method expect a RenderedImage as input, I'm using the following code to create a BufferedImage from the raw image data :

int[] rasterData = new int[width * height];
int rawBufferOffset = 0;
for(int i=0;i<rasterData.length;i++) {
   rasterData[i] = ((int) rawBuffer[rawBufferOffset + 1] << 8) | ((int) rawBuffer[rawBufferOffset] & 0xFF);
   rawBufferOffset += 2;
}
BufferedImage image = new BufferedImage(width,   height,BufferedImage.TYPE_USHORT_GRAY);
image.getRaster().setPixels(0, 0, width, height, rasterData);

The code works but it's obviously not the best method to this conversion, I was thinking about creating a RenderedImage implementation that uses the rawBuffer as the image raster data source, can anyone suggest how to do so or suggest any other method for this conversion?

Answer 1

The most straight forward way, is probably to use a ByteBuffer to swap the byte order, and create a new short array to hold the pixel data.

Then wrap the ( short ) pixel data in a DataBufferUShort . Create a matching WritableRaster and ColorModel , and finally create a BufferedImage from this. This image should be identical to the image in your code above ( BufferedImage.TYPE_USHORT_GRAY ), but be slightly faster to create, as you only copy the pixels once (as opposed to twice in your code).

int w = 2;
int h = 2;
int stride = 1;

byte[] rawBytes = {0x21, 0x27, 0x33, (byte) 0xF6, 0x28, (byte) 0xF3, (byte) 0x27, (byte) 0xF2};
short[] rawShorts = new short[rawBytes.length / 2];

ByteBuffer.wrap(rawBytes)
        .order(ByteOrder.LITTLE_ENDIAN)
        .asShortBuffer()
        .get(rawShorts);

DataBuffer dataBuffer = new DataBufferUShort(rawShorts, rawShorts.length);
WritableRaster raster = Raster.createInterleavedRaster(dataBuffer, w, h, w * stride, stride, new int[]{0}, null);
ColorModel colorModel = new ComponentColorModel(ColorSpace.getInstance(ColorSpace.CS_GRAY), false, false, Transparency.OPAQUE, DataBuffer.TYPE_USHORT);

BufferedImage image = new BufferedImage(colorModel, raster, colorModel.isAlphaPremultiplied(), null);

Another, slightly more convoluted, but probably faster way (as you don't copy the backing pixel array at all), is to create a custom SampleModel that works with MSB (little endian) byte data, but exposes them as TYPE_USHORT . This will create a TYPE_CUSTOM image.

int w = 2, h = 2, stride = 2;
byte[] rawBytes = {0x21, 0x27, 0x33, (byte) 0xF6, 0x28, (byte) 0xF3, (byte) 0x27, (byte) 0xF2};
DataBuffer dataBuffer = new DataBufferByte(rawBytes, rawBytes.length);

SampleModel sampleModel = new ComponentSampleModel(DataBuffer.TYPE_USHORT, w, h, stride, w * stride, new int[] {0}) {
    @Override
    public Object getDataElements(int x, int y, Object obj, DataBuffer data) {
        if ((x < 0) || (y < 0) || (x >= width) || (y >= height)) {
            throw new ArrayIndexOutOfBoundsException("Coordinate out of bounds!");
        }

        // Simplified, as we only support TYPE_USHORT
        int numDataElems = getNumDataElements();
        int pixelOffset = y * scanlineStride + x * pixelStride;

        short[] sdata;

        if (obj == null) {
            sdata = new short[numDataElems];
        }
        else {
            sdata = (short[]) obj;
        }

        for (int i = 0; i < numDataElems; i++) {
            sdata[i] = (short) (data.getElem(bankIndices[i], pixelOffset + bandOffsets[i] + 1) << 8|
                    data.getElem(bankIndices[i], pixelOffset + bandOffsets[i]));
        }

        return sdata;
    }
};
ColorModel colorModel = new ComponentColorModel(ColorSpace.getInstance(ColorSpace.CS_GRAY), false, false, Transparency.OPAQUE, DataBuffer.TYPE_USHORT);
WritableRaster raster = Raster.createWritableRaster(sampleModel, dataBuffer, null);

BufferedImage image = new BufferedImage(colorModel, raster, colorModel.isAlphaPremultiplied(), null);

I don't really see a reason for creating a RenderedImage subclass for this.