Image Processing Reference
the DC coefficient filling the entire area and then the AC coefficients being added back on
top to perturb the pixel values.
Sometimes the frequency domain is referred to as a power spectrum, and hence you
may find the term “energy level” used when discussing the frequency components that
contribute to the image detail in the DCT output graph.
This DCT compression process is applied to the luma and the two chroma pixel
arrays independently of one another.
Phew! That was probably the most difficult part of the process. Everything else is a
lot simpler to understand.
Quantization artifacts as they apply to video are covered in Chapter 5 and more extensively
in Chapter 7 in the discussion about digital audio. Quantizing to a word size that is inad-
equate to resolve all the intensity values causes contours. This is evident on a gradient
shaded area. Figure 10-14 presents an example just to remind you about the effect.
This happens because the word size is trading off resolution in the Z-axis of the
image. The Z-axis controls the intensity of the pixel illumination. Sacrificing resolution in
the Z-axis reduces the range of colors or grayscale shades available to express the image.
The same data-reduction principle is applied to the DCT-coded version of the mac-
roblock. Because that block of pixels has been transformed into a set of frequencies that
describes the image, the loss of detail shows up in a very different way. It is called an arti-
fact, a word that is used very often when compressing video and comparing outputs
Truncating the higher-order harmonic frequencies will lead to a loss of detail in the
image, but the overall appearance of the macroblock remains the same. At an extreme
Figure 10-14 Severely contoured image.