Once the transform or subband coefficients are generated, the goal of stage 2 is to discard and de-emphasize visually insignificant information. This is achieved by quantizing the transform coefficients. We use an efficient procedure for obtaining the optimal subband quantizers. The fundamental idea is that when the subbands are encoded independently, the quantizers should be selected such that all subbands operate at the same level of rate-distortion trade-off. The procedure for finding jointly the optimal rate-distortion trade-off and best basis was proposed by Ramchandran and Vetterli [RV93]. We use the algorithm to find the best space and frequency basis and quantizers. We also use uniform scalar quantizers for the DC subbands and deadzone scalar quantizers for other subbands. The deadzone quantizers work very well in producing runs of zeros which are encoded efficiently in the next stage [VK95].