Compression: JPEG DCT codec

• Usually an 8x8 pixel block is input, representing values of say luminance.
• The forward DCT transforms to an 8x8 array of coefficients for the corresponding frequency representation.
• Quantization uses tables, such as based on the human visual system's sensitivity to each of the 64 frequency components. Thus the first component may be multiplied by 1 to keep with full accuracy, while the last may be multiplied by .001 since it is scarcely perceived.
• Entropy encoding is usually done with Huffman coding, using tables that reflect the frequency of values.
• The resulting compressed bitstream can be stored or transmitted. Eventually it is decoded by being processed through steps that are the inverse of the encoding steps, in reverse order.
• The final result is an 8x8 array that usually has little perceived difference from the original raster image.