The mainstream approach to subband coding has been to partition the in
put signal into subband signals and to code those signals separately w
ith optimal or near-optimal quantizers and entropy coders. A more effe
ctive approach, however, is one where subband coders are optimized joi
ntly so that the average distortion introduced by the subband quantize
rs is minimized subject to a constraint on the output rate of the subb
and encoder. In this paper, a subband coder with jointly optimized mul
tistage residual quantizers and entropy coders is introduced and appli
ed to image coding. The high performance of the coder is attributed to
its ability to exploit statistical dependencies within and across the
subbands. The efficiency of the multistage residual quantization stru
cture and the effectiveness of the statistical modeling algorithm resu
lt in an attractive balance among reproduction quality, rate, and comp
lexity.