An optimal algorithm for exchange between quantization noise and signal spectrum in transform image coding
Hiroshi Ito, Yoshiko Hatano · Electronics and Communications in Japan (Part I Communications) · 1999
In transform coding such as DCT, quantization noise is a function of the input signal when the bit rate is fixed. Hence, some input signals may cause intolerable coding artifacts which appear as blocking effects and mosquito noise in the case of image coding. This paper defines the maximum tolerable level of quantization noise and proposes an algorithm which exchanges the excessive part of the quantization noise with loss of the high-frequency component of the signal. Both bit allocation and signal resolution are given by the algorithm, based on a probability density function (pdf) model of the transform coefficients. The optimality of the algorithm is discussed and a sufficient condition for optimality is derived. This algorithm attains the target quantization noise within the accuracy of the pdf model. The performance is estimated for a test image through computer simulations. © 1999 Scripta Technica, Electron Comm Jpn Pt 1, 82(12): 49–58, 1999