Adaptive vector quantization using reinitialization method

✍ Scribed by Takeshi Nishida; Shuichi Kurogi; Tomonori Saeki

• Book ID: 102823656
• Publisher: John Wiley and Sons
• Year: 2005
• Tongue: English
• Weight: 752 KB
• Volume: 88
• Category: Article
• ISSN: 8756-663X
• DOI: 10.1002/ecjb.20181

✦ Synopsis

In vector quantization (VQ), which is useful for the digital coding of large-volume signals such as for image or audio data, many conventional techniques had assumed that the input signal has a time-invariant probability distribution. However, in a real setting, since the stochastic nature of the environment such as input signal or sensor characteristics vary with time, several VQ techniques that can adapt to these kinds of input signal variations have been proposed in recent years. These techniques have problems such as descending to a local solution or slow adaptation. Therefore, in this paper, the authors use a technique called the reinitialization method to propose an adaptive VQ technique that adapts to input signal changes faster than conventional techniques. In this paper, they first show the reinitialization method for solving the local solution problem of the gradient method and then perform an experiment to compare the performance of conventional techniques and the proposed technique concerning VQ for two-dimensional and higher-dimensional vectors. The results verified that the proposed technique had higher adaptive capabilities than conventional techniques relative to temporal variations of the probability distribution of the input signal.