Image subband coding: a quantization error analysis

Summary form only given. The errors due to quantization in an image subband filter bank have been analyzed by explicitly incorporating a model for a Max-Lloyd quantizer into a quadrature mirror filter (QMF) splitting and reconstruction scheme. The used quantizer model, known as the gain-plus-additive-noise model, consists of a signal part with less than unity gain and a random part that is uncorrelated with the input signal. Measurements show that this model accurately fits the data. Moreover, it enables such coding errors as aliasing, QMF design errors, signal errors, and random errors to be discriminated. A mean-squared error (MSE) calculation and a subjective evaluation of these coding errors on both monochrome and color images have shown that luminance errors are more dominant than chrominance errors, even when the chrominance signals only use 25% of the bits. Further, aliasing errors do occur in the case of quantization. They are not completely cancelled anymore by the QMFs. These errors, however, can be neglected for filter lengths of 16 taps or more. The signal error determines the sharpness of the reconstructed image, while the random error appears in flat areas but also round edges