Impact of dead zone size on the rate/distortion performance of wavelet-based perceptual image encoders

When using uniform quantization schemas with dead zone, the final R/D performance may be affected by two parameters, (a) the dead zone size, and (b) reconstruction point location inside each quantization step. In this work we analyze how the dead zone size affects the image quality for wavelet based encoders that have been perceptually enhanced by using the Contrast Sensitivity Function (CSF). Since the application of CSF may change the distribution of wavelet coefficients in each quantization step, and particularly in the dead zone surroundings, choosing appropriate values for both quantization parameters, dead zone size and reconstruction point, may introduce benefits in terms of R/D performance. After a thorough study about the effects of both parameters, we observed that tuning a variable dead zone quantizer with optimum dead zone and reconstruction point values, rate savings up to 7.7% can be obtained when comparing with popular uniform or fixed dead zone quantizers.