Then group the pixels for each component into 8x8 blocks; transform each block through DCT
-
Note for later -- need blocks as natural frequencies varies over image and not constant on a line of pixels as would correspond to Fourier transforms over full x or y of image. Wavelets "naturally" chooses block size. JPEG has fixed blocks
|
|
In each block, divide each of the 64 frequency components by a separate "quantization coefficient" and round to integers; fundamental lossy step
|
|
Encode the reduced coefficients using Huffman or arithmetic coding (license)
|
|
Add headers and output the result
|
