Abstract: A method and apparatus encodes primary blocks included in an image signal based on the adaptive quantization technique. First, each primary block is divided into first subblocks of M.times.N data elements and second subblocks of P.times.Q data elements, wherein P and Q are multiples of M and N. A covariance for each first subblock is then calculated and compared with a threshold TH1 to thereby determine scalar blocks and primitive vector blocks, wherein a scalar block corresponds to a first subblock having a covariance greater than TH1 and a primitive vector block corresponds to a first subblock having a covariance not greater than TH1. Thereafter, a second subblock containing only primitive vector blocks therein is set as a candidate vector block and a covariance for the candidate vector block is calculated.