engleski

On the statistical properties of the wavelet quantization noise

The discrete wavelet transform has properties to effectively decorrelate and compact energy of digital images. With a quantization scheme based on the human visual characteristic and with an effective coding technique a high compression rate of images is possible. The probability density function (PDF) of the introduced quantization noise depends on the subbands wavelet coefficients statistics and quantization step q. The wavelet coefficients PDF corresponds well with a general Gauss law, and we show that parameter r of that law varies in wide range in dependence of the image contents, level of decomposition and the wavelet used. The quantization error PDF depends strongly on values of r, as well as on sigma/q ratio (sigma is a standard deviation of the wavelet coefficients), and for low r and low sigma/q error PDF is not uniform.

Discrete wavelet transform; uniform quantization; quantization noise

nije evidentirano