Ashdown, Ian, Color Quantization Bibliography, available at <http://liinwww.ira.uka.de/bibliogr aphy/Graphics/cquant.html>

Conversion among types

In Figure 6.1, traversal from left to right corresponds to conversions that can be accomplished without loss.

Disregarding pseudocolour for the moment, data in any of the other three schemes of Figure 6.1 can be “widened” to any scheme to the right simply by assigning the codes appropriately. For example, a greyscale image can be widened to truecolour by assigning codes from black to white. Widening adds bits but not information.

A pseudocolour image can be converted to truecolour through software application of the CLUT. Conversion to truecolour can be accomplished without loss, provided that the truecolour LUTs are sensible.

Concerning conversions in the reverse direction, an image can be “narrowed” without loss only if it contains only the colours or shades available in the mode to its left in Figure 6.1; otherwise, the conversion will involve loss of shades and/or loss of colours.

A truecolour image can be approximated in pseudocolour through software application of a fixed colourmap. Alternatively, a colourmap quantization algorithm can be used to examine a particular image (or sequence of images), and compute a colourmap that is optimized or adapted for that image or sequence.

Image width is the product of socalled resolution and the count of image columns; height is computed similarly from the count of image rows.

Image files

Images in bilevel, greyscale, pseudocolour, or truecolour formats can be stored in files. A general-purpose image file format stores, in its header information, the count of columns and rows of pixels in the image.

Many file formats – such as TIFF and EPS – store information about the intended size of the image. The intended image width and height can be directly stored, in absolute units such as inches or millimeters. Alternatively, the file can store sample density in units of pixels per inch (ppi), or less clearly, dots per inch (dpi). Sample density is often confusingly called “resolution.”

In some software packages, such as Adobe Illustrator, the intended image size coded in a file is respected. In other software, such as Adobe Photoshop, viewing at 100% implies a 1:1 relationship between file pixels and display device pixels, disregarding the number of pixels per inch in the file and at the display. Image files

A point is a unit of distance equal to 1⁄72 inch. The width of the stem of this bold letter I is one point, about 0.353 mm (that is, 353 µm).

without size information are often treated as having 72 pixels per inch; application software unaware of image size information often uses a default of 72 ppi.

“Resolution” in computer graphics

In computer graphics, a pixel is often regarded as an intensity distribution uniformly covering a small square area of the screen. In fixed-pixel displays such as liquid crystal displays (LCDs), plasma display panels (PDPs), and digital light processing (DLP) displays, discrete pixels such as these are constructed on the display device. When such a display is driven digitally at native pixel count, there is a one-to-one relationship between framebuffer pixels and device pixels. However, a graphic subsystem may resample by primitive means when faced with a mismatch between framebuffer pixel count and display device pixel count. If framebuffer count is higher, pixels are dropped; if lower, pixels are replicated. In both instances, image quality suffers.

CRT displays typically have a Gaussian distribution of light from each pixel, as I will discuss in the next chapter. The typical spot size is such that there is some overlap in the distributions of light from adjacent pixels. You might think that overlap between the distributions of light produced by neighboring display elements is undesirable. However, image display requires a certain degree of overlap in order to minimize the visibility of pixel structure or scan-line structure. I will discuss this issue in Image structure, on page 75.

Two disparate measures are referred to as resolution in computing:

•The count of image columns and image rows – that is, columns and rows of pixels – in a framebuffer

•The number of pixels per inch (ppi) intended for image data (often misleadingly denoted dots per inch, dpi)

An image scientist considers resolution to be delivered to the viewer; resolution is properly estimated from information displayed at the display surface (or screen) itself. The two measures above all limit resolution, but neither of them quantifies resolution directly. In Resolution, on page 97, I will describe how the term is used in image science and video.