VIPS from C++

Using VIPS — How to use the VIPS library from C++

Introduction

VIPS comes with a convenient C++ API. It is a very thin wrapper over the C API and adds automatic reference counting, exceptions, operator overloads, and automatic constant expansion. You can drop down to the C API at any point, so all the C API docs also work for C++. 

/* compile with:
 *      g++ -g -Wall try.cc `pkg-config vips-cc --cflags --libs`
 */

#include <vips/vips8>

using namespace vips;

int
main( int argc, char **argv )
{
  GOptionContext *context;
  GOptionGroup *main_group;
  GError *error = NULL;

  if( VIPS_INIT( argv[0] ) )
    vips_error_exit( NULL ); 

  context = g_option_context_new( "" ); 

  main_group = g_option_group_new( NULL, NULL, NULL, NULL, NULL );
  g_option_context_set_main_group( context, main_group );
  g_option_context_add_group( context, vips_get_option_group() );

  if( !g_option_context_parse( context, &argc, &argv, &error ) ) {
    if( error ) {
      fprintf( stderr, "%s\n", error->message );
      g_error_free( error );
    }

    vips_error_exit( NULL );
  }

  VImage in = VImage::new_from_file( argv[1], 
    VImage::option()->
      set( "access", VIPS_ACCESS_SEQUENTIAL_UNBUFFERED ) ); 

  double avg = in.avg(); 

  printf( "avg = %g\n", avg ); 
  printf( "width = %d\n", in.width() ); 

  VImage in = VImage::new_from_file( argv[1],
    VImage::option()->
      set( "access", VIPS_ACCESS_SEQUENTIAL_UNBUFFERED ) ); 

  VImage out = in.embed( 10, 10, 1000, 1000, 
    VImage::option()->
      set( "extend", VIPS_EXTEND_BACKGROUND )->
      set( "background", 128 ) );

  out.write_to_file( argv[2] );

  vips_shutdown();

  return( 0 );
}

Everything before VImage in = VImage::.. is exactly as the C API. This boilerplate gives the example a set of standard command-line flags.

This line is the C++ equivalent of vips_image_new_from_file(). It works in the same way, the differences being:

  • VImage lifetime is managed automatically, like a smart pointer. You don't need to call g_object_unref().

  • Instead of using varargs and a NULL-terminated option list, this function takes an optional VOption pointer. This gives a list of name / value pairs for optional arguments to the function.

    In this case we request unbuffered IO for the image, meaning, we expect to do a single top-to-bottom scan of the image and do not need it to be decompressed entirely.

    The function will delete the VOption pointer for us when it's finished with it.

  • Instead of returning NULL on error, this constructor will raise a VError exception.

There are a series of similar constructors which parallel the other constructors in the C API, see VImage::new_from_memory(), VImage::new_from_buffer(), and VImage::new_matrix(). There's also VImage::new_memory() and VImage::new_temp_file(), which when written to with VImage::write() will create whole images on memory or on disc.

The next line finds the average pixel value, it's the equivalent of the vips_avg() function. The differences from the C API are:

  • VImage::avg() is a member function: the this parameter is the first (the only, in this case) input image.

    The function returns the first output parameter, in this case the average pixel value. Other return values are via pointer arguments, as in the C API.

    Like VImage::new_from_file(), function raises the VError exception on error.

    Like VImage::new_from_file(), extra arguments are passed via an optional VOption parameter. There are none in this case, so the function brackets can be left empty.

All other operations follow the same pattern, for example the C API call vips_add(): 

int vips_add( VipsImage *left, VipsImage *right, VipsImage **out, ... );

appears in C++ as: 

VImage VImage::add( VImage right, VOption *options = 0 );

The next line uses VImage::width() to get the image width in pixels. There are similar functions paralleling vips_image_get_format() and friends. Use VImage::set() to set metadata fields, VImage::get_int() and c. to fetch metadata.

Next we reload the image. The VImage::avg() will have scanned the image and reached the end of the file, we need to scan again for the next operation. If we'd selected random access mode (the default) in the original VImage::new_from_file(), we would not need to reload.

The next line runs vips_embed() with two optional parameters. The first sets the value to an enum (you just use the ones from the C API), the second sets the value to an int. The "background" parameter is actually a VipsArrayDouble: if you pass an int instead, it will be automatically converted to a one-element array for you. You can pass a std::vector<double> too: the utility function VImage::to_vectorv() is a convenient way to make one.

Finally, VImage::write_to_file() will write the new image to the filesystem. You can add a VOption as a final parameter and set options for the writer if you wish. Again, the operation will throw a VError exception on error. The other writers from the C API are also present: you can write to a memory array, to a formatted image in memory, or to another image.

Automatic constant expansion

The C++ API will automatically turn constants into images in some cases. For example, you can join two images together bandwise (the bandwise join of two RGB images would be a six-band image) with: 

VImage rgb = ...; 
VImage six_band = rgb.bandjoin( rgb );

You can also bandjoin a constant, for example: 

VImage rgb_with_alpha = rgb.bandjoin( 255 );

Will add an extra band to an image, with every element in the new band having the value 255. This is quite a general feature. You can use a constant in most places where you can use an image and it will be converted. For example: 

VImage a = (a < 128).ifthenelse( 128, a );

Will set every band element of a less than 128 to 128.

The C++ API includes the usual range of arithmetic operator overloads. You can mix constants, vectors and images freely.

Enum expansion

VIPS operations which implement several functions with a controlling enum, such as vips_math(), are expanded to a set of member functions named after the enum. For example, the C function: 

int vips_math( VipsImage *in, VipsImage **out, VipsOperationMath math, ... );

where VipsOperationMath has the member VIPS_OPERATION_MATH_SIN, has a C convenience function vips_sin(): 

int vips_sin( VipsImage *in, VipsImage **out, ... );

and a C++ member function VImage::sin(): 

VImage VImage::sin( VOption *options = 0 );

Extending the C++ interface

The C++ interface comes in two parts. First, VImage8.h defines a simple layer over GObject for automatic reference counting, then a generic way to call any vips8 operation with VImage::call(), then a few convenience functions, then a set of overloads.

The member function for each operation, for example VImage::add(), is generated by a small Python program called gen-operators.py, and its companion, gen-operators-h.py to generate the headers. If you write a new VIPS operator, you'll need to rerun these programs to make the new member function.

You can write the wrapper yourself, of course, they are very simple. The one for VImage::add() looks like this: 

VImage VImage::add(VImage right, VOption *options)
  throw VError
{
  VImage out;

  call("add" ,
    (options ? options : VImage::option()) ->
      set("out", &out) ->
      set("left", *this) ->
      set("right", right));

  return out;
}

Where VImage::call() is the generic call-a-vips8-operation function.

Generated by GTK-Doc V1.20