JPEG Still Picture Compression Standard
Gregory Wallace

Jonathan Ledlie (jonathan@eecs)

The forward (encoding) and inverse (decoding) Discrete Cosine
Transforms are the basis for JPEG compression.  An image is divided
into 8x8 pixel segments, sent through a DCT codec (encoder and/or
decoder), and then further compressed with either Huffman or
arithmetic coding.  We have encountered this idea of utilizing a
transformation to turn an input into a form more amenable to some
other simple, effecient type of compression in the Burrows-Wheeler
transform.  The output of the forward DCT can be thought of as 64
separate amplitudes, or coefficients.  These coefficients can then be
averaged to form a quantitized DC coeficient, which is useful in
comparing the difference between several adjacent sets of 8x8 pixels.
The 64 DCT coefficients are fed into a 64-element, user-specified
Quantitization table which further compresses the DCT coefficients
with no greater precision than is necessary to achieve the desired
image quality.  The choice of Quantitization table enables different
rates of compression (bits per pixel), given the same DCT
coefficients.  This includes lossless compression, because the DCT
transformation is lossless.