VistaIOImage(3) image representation


#include <vistaio.h>
VistaIOImage image;



Vista represents an image as one or more two-dimensional arrays of pixel values. Each two-dimensional array, called a band, has the same size, which is expressed as a number of rows and a number of columns. Multiple bands may, for example, be used to represent multiple color components, multiple frames of a motion sequence, the real and imaginary components of a complex image, or some combination of these. Besides pixel data, an image may also has an arbitrary list of attributes associated with it.

An image in memory is referred to by the C pointer type VistaIOImage. In data files it's identified by the type name image, and in memory, by the VistaIORepnKind code VistaIOImageRepn. Since it is a standard object type with built-in support in the Vista library, images can be read from data files, written to data files, and manipulated as part of attribute lists by routines such as VistaIOReadFile(3), VistaIOWriteFile(3), and VistaIOGetAttr(3).

Creating and Destroying Images

The following routines create and destroy images.
VistaIOImage VistaIOCreateImage (int nbands, int nrows, int ncolumns, VistaIOPixelRepn pixel_repn)
VistaIOCreateImage allocates memory for a new image with the specified
numbers of bands, rows, and columns. Its pixel values are not initialized, and it is given an empty attribute list. The pixel_repn argument is described below, under ``Pixel Representation''. (See VistaIOCreateImage(3) for details.)

void VistaIODestroyImage (VistaIOImage image)

VistaIODestroyImage releases the memory occupied by image, including that occupied by its pixel data and attribute list.

Image Size

An image's numbers of bands, row, and columns are sometimes abbreviated as nbands, nrows, and ncolumns. The following macros access those numbers.

int VistaIOImageNBands (VistaIOImage image)

int VistaIOImageNRows (VistaIOImage image)

int VistaIOImageNColumns (VistaIOImage image)

VistaIOImageNBands, VistaIOImageNRows, VistaIOImageNColumns return the number of bands, rows, and columns that image has.

int VistaIOImageNPixels (VistaIOImage image)

VistaIOImageNPixels returns the number of pixel values image has altogether (i.e., the product nbands * nrows * ncolumns).

Pixel Representation

An image's pixel values may have any of the following types: VistaIOBit, VistaIOUByte, VistaIOSByte, VistaIOShort, VistaIOLong, VistaIOFloat, or VistaIODouble. VistaIOariables meant to hold pixel values or pointers to pixel values should be defined using these same types (rather than built-in C types such as short and long) to ensure software portability.

The following macros return information about an image's pixel representation:

VistaIORepnKind VistaIOPixelRepn (VistaIOImage image)

VistaIOPixelRepn returns a code indicating image's pixel representation. If image has VistaIOBit pixels, for example, it returns VistaIOBitRepn.

size_t VistaIOPixelSize (VistaIOImage image)

VistaIOPixelSize returns the size, in bytes, used for storing a single pixel of image (i.e, a pixel within a single band, row, and column). This value depends on machine architecture.

int VistaIOPixelPrecision (VistaIOImage image)

VistaIOPixelPrecision returns the minimum size, in bits, needed for representing a single pixel of image. This value does not depend on machine architecture, but only on the kind of pixel value representation used by image. If image has VistaIOBit pixels, for example, this macro returns 1.

VistaIOStringConst VistaIOPixelRepnName (VistaIOImage image)

VistaIOPixelRepnName returns a string describing the type of pixel value representation used by image. For example, if image has VistaIOBit pixels, the string ``bit'' is returned.

VistaIODouble VistaIOPixelMaxValue (VistaIOImage image)

VistaIOPixelMaxValue returns the maximum value that a pixel of image may have. For example, if image has VistaIOSByte pixels, it returns 127.0.

VistaIODouble VistaIOPixelMinValue (VistaIOImage image)

VistaIOPixelMinValue returns the minimum value that a pixel of image may have. For example, if image has VistaIOSByte pixels, it returns -128.0.

size_t VistaIOImageSize (VistaIOImage image)

VistaIOImageSize returns the size, in bytes, of image's block of pixel data, which contains the values of all pixels in all bands. This value depends on machine architecture.

Pixel VistaIOalues

A particular pixel value is indexed by a band number, a row number, and a column number. Bands, rows, and columns are numbered consecutively from zero. Rows number down and columns number to the right so that row 0, column 0 is at the upper-left corner of the image.
type VistaIOPixel (VistaIOImage image, int band, int row, int column, type)
VistaIOPixel accesses the pixel at a specified band, row and column in an
image whose pixels are of a specified type. The pixel type, type, must be one of VistaIOBit, VistaIOUByte, VistaIOSByte, VistaIOShort, VistaIOLong, VistaIOFloat, or VistaIODouble. The macro may be used either as an rvalue (to fetch the pixel's value) or as an lvalue (to modify it or obtain its address). The ``EXAMPLES'' section, below, demonstrates its use as an rvalue.
VistaIOPointer VistaIOPixelPtr (VistaIOImage image, int band, int row, int column)

VistaIOPixelPtr returns the address of a pixel specified by its
band, row, and column coordinates. The pixel at that address can be fetched or modified by first coercing the pointer to the appropriate type for the pixel. For example:

pixel = VistaIOPixelPtr (image, band, row, column);
if (VistaIOPixelRepn (image) == VistaIOBitRepn)
* (VistaIOBit *) pixel = new_value;

type ** VistaIOPixelArray (VistaIOImage image, type)

VistaIOPixelArray returns a pointer that can be used to access any image pixel by indexing it first with the pixel's band number, then its row number, and then its column number. For example:

VistaIOUByte **pixels = VistaIOPixelArray (image, VistaIOUByte);
pixels[band][row][column] = new_value;

VistaIOPointer VistaIOImageData (VistaIOImage image)

VistaIOImageData returns a pointer to image's block of pixel data. The block has the size VistaIOImageSize (image) bytes.

Other Image Attributes

In addition to its pixel values an image may have any number of other attributes; they are represented as an attribute list.

VistaIOAttrList VistaIOImageAttrList (VistaIOImage image)

VistaIOImageAttrList is a macro for accessing the attribute list associated with an image. The macro may be used as an rvalue to reach attributes within the list:

VistaIOGetAttr (VistaIOImageAttrList (image), VistaIONameAttr, ...)

or as an lvalue to manipulate the entire list:

VistaIODestroyAttrList (VistaIOImageAttrList (image));
VistaIOImageAttrList (image) = VistaIOCopyAttrList (other_list);

Band Interpretation

The bands of a multi-band image might represent the successive frames of a motion sequence, the left and right halves of a stereo pair, or the red, green, and blue channels of a color image. They might even represent a combination of such dimensions --- for example, a stereo pair of color images. Because a great many different uses of the bands are possible, each image includes information describing how its bands are to be interpreted. This information assigns each band an interpretation in terms of four dimensions:
covers the various frames of a motion sequence
covers, for example, the left and right channels of a stereo pair
covers, for example, the red, green, and blue channels of a color image
covers, for example, the real and imaginary components of a complex image

Although these four dimensions do not account for all conceivable ways one might wish to use the bands of a multi-band image, they do cover most cases. The scheme can usually be adapted to cover cases not explicitly handled. For example, a collection of images obtained under various lighting conditions could be represented by using the frame dimension to index the lighting condition.

Each image specifies the sizes of its four band interpretation dimensions. For a stereo pair of color images, for example, the size of the viewpoint dimension is two, the size of the color dimension is three, and the remaining dimensions have sizes of one. The dimension sizes are often abbreviated as nframes, nviewpoints, ncolors, and ncomponents. Their product should always equal the number of bands in the image. These macros exist for accessing the sizes:

int VistaIOImageNFrames (VistaIOImage image)

int VistaIOImageNViewpoints (VistaIOImage image)

int VistaIOImageNColors (VistaIOImage image)

int VistaIOImageNComponents (VistaIOImage image)

Each macro may be used as rvalue to obtain the size of a dimension, or as an lvalue to set it.

The four dimensions are mapped onto the single dimension, band, according to a particular ordering: component varies most rapidly, then color, then viewpoint, and finally frame. A macro is available for computing this mapping:

int VistaIOBandIndex (VistaIOImage
image, int frame, int viewpoint, int color, int component)

VistaIOBandIndex returns the index of the band that corresponds to a particular combination of frame, viewpoint, channel, and component indices.

An image's attribute list may include attributes indicating how the image employs the four band interpretation dimensions. One attribute for each dimension provides the first level of description. Its value is an integer code, with some standard codes defined for common uses. Below is a list of the four attributes and their standard values. Shown in are both the symbolic constants that can be used C programs to refer to the integer values, and (in parentheses) the keywords used to represent the values externally, in Vista data files.

frame_interp, which currently has no standard values defined for it

viewpoint_interp, which has this standard value:

VistaIOBandInterpStereoPair (stereo_pair)
The two viewpoints represent (respectively) the left and right images of a stereo pair.

color_interp, which has these standard values:

VistaIOBandInterpRGB (rgb)
The three colors represent (respectively) red, green, and blue color channels.

component_interp, which has these standard values:

VistaIOBandInterpComplex (complex)
The two components represent (respectively) the real and imaginary components of a complex image.
VistaIOBandInterpGradient (gradient)
There is a single component representing gradient magnitude, two components representing (respectively) the x (rightward) and y (upward) components of gradient, or three representing the x, y, and z (forward in band sequence) components of gradient.
VistaIOBandInterpIntensity (intensity)
There is a single component representing image intensity. For integer pixel representations, pixel values normally span the range between zero and the maximum representable pixel values (e.g., [0,127] for VistaIOSByte pixels). For floating point pixel representations, pixel values normally span the range [-1,+1].
VistaIOBandInterpOrientation (orientation)
There is a single component representing orientation. Pixel values express orientation in radians, counterclockwise from the direction of increasing column number (rightward).

To prevent accidental misspellings of the attribute names you should use these symbolic constants:

#define VistaIOFrameInterpAttr "frame_interp"
#define VistaIOViewpointInterpAttr "viewpoint_interp"
#define VistaIOColorInterpAttr "color_interp"
#define VistaIOComponentInterpAttr "component_interp"

A dictionary, VistaIOBandInterpDict, defines the mapping between the band interpretation codes (e.g., VistaIOBandInterpComplex) and their keywords (e.g., complex):

VistaIODictEntry VistaIOBandInterpDict[];

For each of the four band interpretation dimensions there is a routine you can use to simultaneously check both the dimension's size and, if present, its interpretation attribute. The VistaIOBandInterp(3) man page describes the four routines.

Pixel Aspect Ratio

Pixel aspect ratio is defined as the ratio of pixel width to pixel height. An image may include, on its attribute list, an attribute specifying this ratio as a floating point value. The attribute has the name pixel_aspect_ratio and a symbolic constant exists for referring to it:

#define VistaIOPixelAspectRatioAttr "pixel_aspect_ratio"

When the attribute is absent, a pixel aspect ratio of 1.0 is assumed.

Calling Conventions

A Vista library routine that processes one image to produce another usually takes three or more arguments, like this:
VistaIOImage VistaIOProcessImage (VistaIOImage src, VistaIOImage dest, VistaIOBand band)

The first argument specifies the source image to be processed. The second specifies the destination image, to contain the result. You generally have three choices for how this destination image is provided:
you can have the routine create a destination image. If you pass NULL for dest, the routine will create and return a destination image that is appropriate for the source image and for the type of operation being performed on it:

dest = VistaIOProcessImage (src, NULL, band);

you can supply a destination image provided it has the correct properties (usually, an appropriate size and pixel representation):

dest = VistaIOCreateImageLike (src);
VistaIOProcessImage (src, dest, band);
you can specify the same image as both source and destination, in which case the source pixel values will be replaced by destination pixel values. This choice is only available for operations where the source and destination images have the same size and pixel representation.

VistaIOProcessImage (src, src, band);

In all cases, the routine will return the destination image if successful, and NULL otherwise. Moreover, if NULL is returned, the routine will already have called VistaIOWarning(3) to issue a warning message describing the problem.

The third argument to most image processing routines, band, may specify a particular band of the source image to be processed; or it may be the constant VistaIOAllBands (defined as -1) to indicate that all bands of the source image are to be processed. When a particular band is specified, band must be at least 0 and less than the number of bands in the source image; the destination image will usually have a single band to contain the result. When VistaIOAllBands is specified, the destination image will usually have the same number of bands as the source image.

Normally a routine will copy the source image's attribute list to the destination image, replacing any attributes the destination image had already. The sizes of the four band interpretation dimensions (nframes, nviewpoints, etc.) are also copied. When appropriate, however, a routine may delete or modify the attributes and dimensional sizes as it copies them from source to destination. For example, a routine that converts an RGB color image to a grey-scale one (VistaIORGBImageToGray(3)) expects a source image with the properties ncolors: 3 and color_interp: rgb; the destination image it produces has ncolors: 1 and no color_interp attribute.

Image Representation in Memory

typedef struct {
int nbands; /* number of bands */ int nrows; /* number of rows */ int ncolumns; /* number of columns */ VistaIORepnKind pixel_repn;/* representation of pixel values */ unsigned long flags; /* various flags */ VistaIOAttrList attributes;/* list of other image attributes */ VistaIOPointer data; /* array of image pixel values */ VistaIOPointer *row_index;/* ptr to first pixel of each row */ VistaIOPointer *band_index;/* ptr to first row of each band */ int nframes; /* number of motion frames */ int nviewpoints; /* number of camera viewpoints */ int ncolors; /* number of color channels */ int ncomponents; /* number of vector components */
} VistaIOImageRec, *VistaIOImage;

An image is represented in memory by a value of type VistaIOImage, which points to a VistaIOImageRec structure. Programmers using images will usually not need to access this structure's fields directly from their code since there are Vista library routines and macros available for most purposes. However, when debugging, one may occasionally need to examine a VistaIOImageRec structure's fields directly. The following paragraphs describe the fields.

The nbands, nrows, and ncolumns fields of the structure describe how many bands, rows, and columns the image has. Its pixel_repn field specifies how the image's pixel values are represented; it contains one of the constants VistaIOBitRepn, VistaIOUByteRepn, VistaIOSByteRepn, VistaIOShortRepn, VistaIOLongRepn, VistaIOFloatRepn, or VistaIODoubleRepn. Its attributes field contains the list of attributes associated with the image.

An image's pixel values are stored in a contiguous block of memory where they are arranged as follows. Within each band pixel values are ordered, beginning with the pixel at row 0, column 0, proceeding to row 0, column 1, etc., and ending with the highest numbered row and column. Each band's array of pixel values occupies a separate block of memory, and these blocks are arranged contiguously. All of band 0's pixel values are first, followed by all of band 1's, etc. In summary, pixel values are ordered by band, then by row within each band, and then by column. The pixel values themselves are of type VistaIOBit, VistaIOUByte, ..., or VistaIODouble.

The data field of a VistaIOImageRec points to the block of memory containing pixel values. The row_index field points to a vector of pointers to the first pixel value of each row within each band; this vector is of length (nbands * nrows) and it is indexed by an expression of the form (band * nrows + row). The band_index field points to a vector of pointers to the first row_index entry of each band; this vector is of length nbands and it is indexed by band number.

Individual bits of a VistaIOImageRec's flags field are used to denote various things. One bit is currently defined:

This bit, if nonzero, indicates that the VistaIOImageRec, the block of pixel values, the row index, and the band index were all obtained with a single call to VistaIOMalloc(3). If the bit is zero, these things were allocated separately and therefore they must be released separately.

The nframes, nviewpoints, ncolors, and ncomponents fields describe the number of motion frames, camera positions, color channels, and vector components that the image's various bands represent. Together they should account for all bands of the image --- i.e., their product should equal nbands. The image's bands are ordered according to these fields, with component index incrementing most rapidly and frame index incrementing least rapidly.

Image Representation in a File

attribute-name: image {
nbands: nbands nrows: nrows ncolumns: ncolumns repn: pixel-representation nframes: nframes nviewpoints: nviewpoints ncolors: ncolors ncomponents: ncomponents data: data-offset length: data-length other attributes
} pixel-representation ::=
bit | ubyte | sbyte | short | long | float | double

An image is represented in a Vista data file as an attribute. The attribute's value is an object of type image containing the following attributes in any order:

an integer-valued attribute specifying the number of bands in the image.
an integer-valued attribute specifying the number of rows in the image.
an integer-valued attribute specifying the number of columns in the image.
a string-valued attribute specifying the image's pixel value representation with one of the following keywords: bit, ubyte, sbyte, short, long, float, or double.
an integer-valued attribute specifying the number of motion frames represented by the image's bands.
an integer-valued attribute specifying the number of camera viewpoints represented by the image's bands.
an integer-valued attribute specifying the number of color channels represented by the image's bands.
an integer-valued attribute specifying the number of vector components represented by the image's bands.
an integer-valued attribute specifying the offset of the image's pixel data within the binary data section of the file
an integer-valued attribute specifying the size, in bytes, of the image's pixel data

An nbands, nframes, nviewpoints, ncolors, or ncomponents attribute may be omitted if its value is 1. Moreover, the image object's attribute list may contain any other attributes associated with the image. See, for example, the data file listed under ``EXAMPLES'', below.

Pixel Representation in a File

Image pixel values are represented in binary form in the binary data section of a file. They occupy a contiguous block in which pixel values are ordered by band, row, and column, as in memory.

Single-bit pixel values are packed eight to a byte, beginning with the most-significant bit of each byte. Pixel values of more than 8 bits are packed into multiple bytes, beginning with the most-significant byte of the pixel value. Consecutive pixel values are packed together, without intervening bits for padding or alignment. If necessary, however, the entire block of pixel values is padded so that its length is a multiple of 8 bits.

Each kind of pixel is represented as follows in the Vista data file:

as a 1-bit unsigned integer
as an 8-bit unsigned integer
as an 8-bit two's-complement integer
as a 16-bit two's-complement integer
as a 32-bit two's-complement integer
as a 32-bit IEEE floating-point number
as a 64-bit IEEE floating point number

Note that a pixel's representation in a Vista data file is independent of the form it takes when stored in memory on any particular machine. Thus a VistaIOBit pixel may occupy 8 bits in the main memory of a Sun SPARCstation 32 bits on a Cray machine, but it always occupies a single bit in a Vista data file.


This code fragment sets all pixels with 1 in a one-band image of single-bit pixels:

for (i = 0; i < VistaIOImageNRows (image); i++)
for (j = 0; j < VistaIOImageNColumns (image); j++)
VistaIOPixel (image, 0, i, j, VistaIOBit) = 1;

The previous example may be made more efficient by avoiding the repeated computation of pixel addresses:

VistaIOBit *p = & VistaIOPixel (image, 0, 0, 0, VistaIOBit);

for (i = 0; i < VistaIOImageNPixels (image); i++)

*p++ = 1;

This code fragment creates an image of complex pixel values:

image = VistaIOCreateImage (2, 256, 256, VistaIOFloatRepn);
VistaIOImageNFrames (image) = VistaIOImageNViewpoints (image) = VistaIOImageNColors (image) = 1;

VistaIOImageNComponents (image) = 2;

VistaIOSetAttr (VistaIOImageAttrList (image), VistaIOComponentInterpAttr, NULL, VistaIOLongRepn, (VistaIOLong) VistaIOBandInterpComplex);

Here is an example of a Vista data file containing two images:

V-data 2 {
one: image {
nbands: 2 nrows: 256 ncolumns: 256 ncomponents: 2 component_interp: complex repn: float data:
offset of first image's pixel values length: 524288 name: "UFO sighted over VistaIOancouver" pixel_aspect_ratio: 1.25
} two: image {
nrows: 32 ncolumns: 32 repn: ubyte data: offset of second image's pixel values length: 128 name: "UFO icon"
} } ^L first image's pixel values second image's pixel values


Art Pope <[email protected]>

Adaption to vistaio: Gert Wollny <[email protected]>


The following table summarizes the Vista library routines that operate on images. Many of these routine are documented elsewhere, by a section 3 man page named for the routine.

For creating and destroying images in memory:

Create an image.
Create one image like another.
Release memory occupied by an image.

For fetching and storing pixel values:

Fetch a pixel value with any pixel representation.
Store a pixel value with any pixel representation.
Select bands of image pixels for processing.
Select a destination for an image processing operation.

For getting band interpretation information:

Report meaning of frame dimension.
Report meaning of color dimension.
Report meaning of component dimension.
Report meaning of viewpoint dimension.

For reading and writing images:

Read a set of images from a data file.
Read an image in Vista plain file format.
Read an image in Portable Anymap (PNM) format.
Read an image in UBC image file format.
Write a set of images to a data file.
Write an image in UBC image file format.

For comparing images:

Test whether two images have the same size and pixel representation.
Test whether two images have the same size.

For copying images:

Copy selected bands of pixel values from various source images to a single destination image.
Copy one or all bands of pixel values from one image to another.
Copy one image to another.
Copy one image's attributes to another.
Copy one image's pixel values to another.

For converting an image's pixel representation:

Convert an image's pixel representation while simply copying pixel values.
Convert an image's pixel representation using some linear mapping of pixel values.
Convert an image's pixel representation while mapping the actual range of source pixel values to the full range of possible destination pixel values.
Convert an image's pixel representation while mapping the full range of possible source pixel values to the full range of possible destination pixel values.

For rotating or transposing an image:

Flip an image horizontally or vertically.
Rotate an image by any angle.
Transpose the rows and columns of an image.

For changing the size of an image:

Extract a rectangular region from an image.
Increase the size of an image by an integer factor.
Decrease the size of an image by an integer factor.
Scale the size of an image, up or down, by any factor.

For filtering and transforming images:

Adjust image brightness and/or contrast.
Detect edges in an image using a Canny operator.
Convolve an image with a mask.
Convolve an image with a Gaussian filter or its derivative.
Compute a forward or inverse Fourier transform.
Compute the horizontal and vertical components of image gradient.
Perform an arithmetic or logical operation on an image's pixel values.
Invert an image's pixel values, swapping black and white.
Create an edge set by linking connected, non-zero image pixels.
Negate an image's pixel values.
Mark the zero crossings in an image.

For computing image statistics:

Compute the minimum, maximum, mean, and variance of an image's pixel values.

For working with complex images:

Build a complex image from separate images of real and imaginary components.
Compute the magnitude of a complex image.
Compute the phase of a complex image, or the gradient direction of a two-component gradient image.

For displaying or rendering images:

Render an image band using PostScript.
Render an RGB color image using PostScript.
An X Toolkit widget for displaying an image.
Compute appropriate window dimensions for displaying an image.

For filling images with patterns:

Fill an image with a constant pixel value.
Fill an image with an intensity ramp.
Fill an image with a sine grating.
Fill an image with a zone plate pattern.
Fill an image with a random mixture of two pixel values.
Fill an image with pixel values drawn from a normal distribution.
Fill an image with pixel values drawn from a uniform distribution.


compute optical flow by weighted least squares.