Routines for loading/decoding and writing/encoding images.

Implemented decoders:

• JPEG: baseline 8-bit (Huffman sequential DCT with 8-bits per channel).
• PNG: loads all images in the PNG test suite.

Usage - loading from a file:

Image img = load("imagepath/imagefile.png");

Usage - loading from a stream:

Stream dataStream;
Decoder dcoder = getDecoder("imagepath/imagefile.png");
while(dcoder.parseStream(dataStream))
{
  do stuff, like draw the current interlaced png, fill the stream, etc.
} 

Implemented encoders:

• PNG: will write out PNG files from 8-bit versions of Image class. Uses adaptive filtering, output PNG's are non-interlaced and only contain critical chunks.

Usage - writing out a PNG:

ubyte[] data = myImageData;  // note that it must be pixel interleaved for RGB/RGBA to work
Image myImg = new Img!(Px.R8G8B8)(width, height, data);
myImg.write("path/to/output.png");

Images:

• the Image class can be used to hold various pixel formats. It also has routines for resizing, e.g.:

Image myImg; 
myImg.resize(newWidth, newHeight, Image.ResizeAlgo.BILINEAR);

• Resizing can be done by cropping (Image.ResizeAlgo.CROP), nearest neighbour (Image.ResizeAlgo.NEAREST) or bilinear filtering (Image.ResizeAlgo.BILINEAR).

simpledisplay.d is from Adam Ruppe's repo: misc-stuff-including-D-programming-language-web-stuff.

Some details:

• JPEG: the decoder uses Nearest Neighbour upsampling of the chroma components by default, but bilinear upsampling is also available. The decoder only retains enough info for one MCU at a time, so it decodes on the fly. Not thoroughly tested, particularly not on greyscale images, so these might not work.

• PNG: better tested thanks to the PNG test suite. Handles all bit depths, but note that sub-byte packing (like 1, 2, 4 bits per pixel) are unpacked when stored in the Image class. This makes them much easier to deal with later, at the expense of increasing their memory footprint. 16 bit resolution is retained, but by default, when reading pixels out of a 16-bit Image, they are downshifted to 8-bit precision, since usually that is what you want. Decoder retains two scanlines of info, so it decodes on the fly and can be used with interlaced PNG's to give a progressive display. Note that this makes it slightly less efficient than it could be. It only decodes critical chunks. Ancillary chunks are easy enough to add though.