IIO
===


1. OVERVIEW
-----------

This is a collection of C functions (not a library), for reading small images
in many formats.

To use the functions, copy the "iio.c" and "iio.h" files into your source code.
Edit the macros at the start of "iio.c" to fine-tune your library requirements.



2. PHILOSOPHY
-------------


An image is an array of numbers.  Not intensities, not luminances, just
numbers.  The iio functions allow you to read these numbers from files.
If the file contains the number 7, then iio will give you 7.  All metadata
regarding the meaning and scaling of these numbers is energically ignored.

Images are always read as a whole, and are copied several times internally.
Thus IIO is only suitable for reading small images.  If you need to read small
portions of large images in an efficient way, take a look at "fancy_image.c"

As a matter of principle, no image processing is ever done by iio.



2. BASIC USAGE
--------------

Read the pixels of "lena.png" into an array of floats:

	int w, h, pd;  // image width, heigth, pixel dimension
	float *x = iio_read_image_float_vec("lena.png", &w, &h, &pd);

Write an array of WxHxD floats into file "out.tif":

	iio_write_image_float_vec("out.tif", x, W, H, D);

When reading an image, the filename extension is ignored.
When writing an image, the extension is used to select the file format.

Reading from the image "-" is equivalent to reading from standard input.
Writing into the image "-" is equivalent to writing to standard output.



3. ADVANCED USAGE
-----------------

3.1. ALTERNATIVE PIXEL ORDERING ( RGBRGBRGB... vs. RRR...GGG...BBB... )

An image is an array of pixels, and each pixel is an array of samples.
This means that for a typical RGB image the numbers are stored as RGBRGBRGB...

If you want RRRRR...GGGG...BBBB... you do not want an image, but an array of
images, where the components of each pixel have been broken into separate
images.  Then you use the "_split" functions instead of the "_vec" functions
from iio.

Read "lena.png" into three contiguous images with the R,G,B components of each
pixel:

	int w, h, pd;  // image width, heigth, pixel dimension
	float *x = iio_read_image_float_split("lena.png", &w, &h, &pd);


3.2. GRAY IMAGES

If you want to work with gray images (e.g., one-dimensional pixels), there are
functions where pd=1 is implicit.  The following code reads a gray-level lena:

	int w, h;  // image width, height
	float *x = iio_read_image_float("lena.png", &w, &h);


3.3. OTHER DATA TYPES

The recommended data type for iio is "float".  However, the other C types are
possible.  For example, the following code quantizes an image to 8 bits

	// quantize a floating-point tiff into a 8-bit png
	int w, h;
	uint8_t *x = iio_read_image_uint8("in.tiff", &w, &h);
	iio_write_image_uint8("out.png", x, w, h);

Notice that, since IIO does not do image processing, the quantized values are
not scaled.  In this example, if "in.tiff" had floating-point values between 0
and 1, then "out.png" will be identically zero.  If you want any kind of value
scaling you have to do it yourself:

	// quantize a floating-point tiff into a 8-bit png
	int w, h;
	float *x = iio_read_image_float("in.tiff", &w, &h);
	uint8_t *y = malloc(w*h);
	for (int i = 0; i < w*h; i++)
		y[i] = 255 * x[i];
	iio_write_image_uint8("out.png", y, w, h);

When converting floats into ints, numeric values outside the range of the int
are clipped, and NANs are converted to 0.


4. MAGIC USAGE
--------------

4.1. MAGIC FILENAMES

Some strings are recognized by IIO as "magic" filenames

	"-"                       standard input (when reading)
	"-"                       standard output (when writing)
	"zero:512x512"            a zero gray image of the indicated size
	"zero:512x512x3"          a zero color image of the indicated size
	"one:WxH"                 an image of ones
	"constant:17:256x256"     a constant gray image = 17
	"constant:nan:256x256x3"  a rgb image full of NANs
	"http://URL"              dowload the image using wget and load it


4.2. MAGIC ACCESSORS

The different parts of a multi-part TIFF image can be accessed by a suffix:

	"file.tiff"               read the first image in the file
	"file.tiff,0"             equivalent to "file.tiff"
	"file.tiff,3"             read the fourth image in the file

Similarly, for the different (named) datasets in an HDF5 image:

	"file.nc"                 read the "/dset" dataset (unlikely to exist)
	"file.nc,/PROD/radiance"  read the named dataset


4.3. RAW IMAGES

Idea: to read a raw file named "file.xxx", open the image
with name "RAW[...]:file.xxx".  The "..." specify the
details of the raw format.


Example:

	RAW[w320,h200,tFLOAT]:file.xxx

This reads 320x200 floats from "file.xxx".

The contents of [ ] are a list of "tokens", separated by ","

Each token is a character followed by its value

Valid characters with their meaning:

	w = width
	h = height
	p = pixel dimension (e.g. 1 or 3)

	t =  one of "INT8", "UINT8", "INT16", "UINT16", "INT32", "UINT32",
	"INT64", "UINT64", "FLOAT", "DOUBLE", "LONGDOUBLE", "HALF", "UINT1",
	"UINT2", "UINT4", "CHAR", "SHORT", "INT", "LONG", "LONGLONG",

	o = offset bytes to be ignored from the start of the file
	    (if negative, ignored from the byte after th end of the file)
	    ((default=-1 == EOF))

	b = 0,1 wether pixel channels are contiguous or broken into planes

	e = 0,1 controls the endianness.  By default, the native one

	r = xy, xY, Xy, XY, yx, Yx, yX, YX
        controls the orientation of the coordinates (uppercase=reverse)

All the numeric fields can be read from the same file.  For example,
"w@44/2" says that the width is read from position 44 of the file
as an uint16, etc.

More magic: Specifying the RAW type using environment variables.

Typically, when IIO fails to recognize the filetype, it tries some desperate
measures.  One of this desperate measures is enabled by the environment
variable  IIO_RAW, that specifies the "rawstring" of the file to be opened.



5. BLACK MAGIC USAGE
--------------------

The suffix "TRANS" allows to apply some transformations to images after being
read.  For example

	TRANS[x=10]:file.png         trim the first 10 columns of the image
	TRANS[y=10]:file.png         trim the first 10 lines of the image
	TRANS[w=100]:file.png        force the image to have width 100
	TRANS[h=100]:file.png        force the image to have height 100
	TRANS[y=10,h=100]:file.png   extract lines 10...110
	TRANS[h=100,y=10]:file.png   extract lines 10...90
	TRANS[flip=topdown]:file.png flip the image top down

Notice that arguments of TRANS are comma-separated and applied sequentially.
Besides "topdown", the "flip" transform admits the following arguments, with
obvious meanings.

	identity
	leftright
	topdown
	transpose
	posetrans
	r90
	r270
	r180

There is a further, very dangerous, argument of TRANS that is named "pipe".  It
pipes image through the indicated shell command.

	TRANS[pipe="qauto -i"]:file.png
		Normalize the colors of the image

	TRANS[pipe="plambda 'split rot rgb'|qauto -i"]:file.png
		Swap the blue and green channels, and normalize

	TRANS[pipe="plambda split_rot_rgb|qauto -i"]:file.png
		Same as before, but easier to escape

Even worse (!):  You can define an environment variable "IIO_TRANS" and the
transformation will be applied to *all* iio-using programs that see the
environement variable.  For example, if you have a folder with a lot of images,
you can browse their fft using pvflip as follows:

	IIO_TRANS="pipe=fft|plambda vnorm_1_+_log" pvflip *.png

That's enough black magic for now, but rest assured that there's more.



6. COMPILE TIME CONFIGURATION
-----------------------------

Edit the definitions at the beginning of file iio.c to fine-tune your
requirements.

These flags can also be set as compiler options,
for example, -DIIO_DISABLE_IMGLIBS


//#define IIO_ABORT_ON_ERROR
//
// Whether to ABORT the program on error or return NULL.

//#define IIO_SHOW_DEBUG_MESSAGES
//
// Wether you want to be extremely verbose (useful for debugging only)

//#define IIO_DISABLE_IMGLIBS
//
// Disable all readers requiring external libraries.

You can also select a particular set of libraries from the compiler command
line.  For example, the options "-DI_CAN_HAS_LIBTIFF -DDISABLE_LIBPNG" enables
TIFF and disables PNG.





7. SUPPORTED FORMATS FOR READING
--------------------------------

The following image formats can be read natively, regardless of the
compile-time options:

	* PNM (any format recognized by netpbm)
	* NUMPY (python's numpy arrays)
	* RIM (megawave's image format)
	* ASC (cimg general 4d arrays)
	* PFM (portable floatmap)
	* QNM (a straightforward generalization of PNM)
	* CSV (a csv array is read into a gray image)
	* VRT (gdal's virtual images)
	* PCM (for optical flows and complex images)
	* FLO (for middlebury optical flows)
	* JUV (optical flows)
	* LUM ("luminance" images)
	* FARBFELD (suckless 16bit photo format)
	* PDS (Planetary Data System)
	* VICAR (astronomical images)
	* CCSD3ZF (PDS with a different magic number)
	* FITS (astronomical images)
	* raw uncompressed image data (see section 4 "MAGIC USAGE" above)

The following formats can be read using external library support (when the
appropriate compile-time option is enabled):

	* PNG
	* JPEG
	* TIFF
	* EXR (disabled by default)
	* HDF5 (disabled by default)



8. SUPPORTED FORMATS FOR WRITING
--------------------------------

Although the main concern of IIO is to READ images, the library
provides some convenience functions for writing images.  This is a
secondary afterthought, and there is no intention to support a
complete set of formats.  Specifically, lossy compression is not
implemented.

By default, images are now saved as (uncompressed) numpy arrays.  If a file
extension is recognized then a different format is used:

	.npy  : Numpy array
	.asc  : Cimg's 4-dimensional array
	.mw   : Megawave's RIM files
	.pgm  : portable graymap
	.ppm  : portable pixmap
	.pfm  : portable floatmap
	.csv  : comma-separated values printed in ascii
	.flo  : a middlebury optical flow
	.uv   : javier's optical flow format
	.six  : sixels (useful for plotting into xterm)

	.tif  :
	.tiff :
	.TIF  :
	.TIFF : tiff format (type is determined by C data type)
	.png  :
	.PNG  : png format (8 bits)

To force TIFF output format, prefix the filename with "TIFF:".  For example,
"TIFF:-" writes a TIFF file to stdout.

To force 16 bits png, use the prefix "PNG16:".



8. FILES
--------

iio.h                     interface (main source of documentation)
iio.c                     implementation

doc/*                     documentation (mostly philosohpical)
samples/*                 test images
mini/iio.{c,h}            mini version, crippled, without deps
contrib/octave/*          octave bindings
contrib/lua/*             lua bindings
contrib/python/*          python bindings
contrib/cmake/*           cmake helper files