- the image is divided in blocks of 8x8 pixels, to which a DCT is applied (the blocks are read line by line, from top to bottom and from left to right). The basic process implies that the levels associated with each pixel are 8 bit coded.
- the 64 coefficients of DCT are quantified (rounded);
- the 'mean value' (DCT value at the frequency 0) is subtracted from the same term of the next block;
- the 63 other terms are read in 'zigzags';
- the sequence of the obtained values is coded (Huffman entropic coding);
- each non-zero coefficient is coded by the number of zeros preceding it, the number of bits needed for its coding, and its value.