Optimum partition of frequency bands in 2-dimensional sub-band image coding
✍ Scribed by Masashi Kameda; Kohhei Ohtake; Makoto Miyahara
- Publisher
- John Wiley and Sons
- Year
- 1998
- Tongue
- English
- Weight
- 277 KB
- Volume
- 81
- Category
- Article
- ISSN
- 1042-0967
No coin nor oath required. For personal study only.
✦ Synopsis
This paper aims to attain high-efficiency sub-band coding of image signals. We present a theoretical study concerning the optimum partition of the 2-dimensional frequency domain obtained by matching the statistical properties, i.e., the auto-correlation function. Initially, a theoretical model is proposed for an arbitrary 2-dimensional band partition and the corresponding filter bank structure. The relation between the sub-band of the analysis filter bank and the output rate is analyzed based on this model. We assume that each sub-band signal after partition is scalar-quantized and present a theoretical expression that determines the optimum band partition and the bit assignment to the sub-band on the 2-dimensional frequency domain, under the following conditions. (1) The bit rate that is assigned to the input signal is held constant. (2) The quantization noise power in the signal reconstructed at the receiving end is minimized. We present an algorithm for numerical computation in order to derive exactly the optimum partition and the optimum bit assignment. Finally, a 2-dimensional first-order Markov signal sequence whose correlation coefficient between adjacent pixels is close to unity is assumed. The result of numerical computation for the optimum partition and the optimum bit assignment based on the proposed theory are compared to those obtained by the conventional band partition method and the effectiveness of the proposed method is quantitatively verified.