Context modeling based on context quantization with application in wavelet image coding
Context modeling based on context quantization with application in wavelet image coding
复制标题
DOI：
10.1109/tip.2003.819224
复制
发表时间：
2004
期刊：
影响因子：
--
通讯作者：
Jianhua Chen
中科院分区：
文献类型：
--
作者：
Jianhua Chen
上下文建模广泛应用于图像编码中以提高压缩性能。然而，如果不进行特殊处理，预期的压缩增益将被高阶上下文模型引入的模型成本所抵消。上下文量化是处理这个问题的有效方法。在本文中，我们详细分析了一般的上下文量化问题，并表明上下文量化与常见的矢量量化问题类似。如果定义了合适的失真测量，则可以通过劳埃德式迭代算法来设计最佳上下文量化器。该上下文量化策略应用于嵌入式小波编码方案，其中使用所提出的量化算法设计的上下文模型通过算术编码直接对重要性图符号和符号符号进行编码。实现了良好的编码性能。
Context modeling is widely used in image coding to improve the compression performance. However, with no special treatment, the expected compression gain will be cancelled by the model cost introduced by high order context models. Context quantization is an efficient method to deal with this problem. In this paper, we analyze the general context quantization problem in detail and show that context quantization is similar to a common vector quantization problem. If a suitable distortion measure is defined, the optimal context quantizer can be designed by a Lloyd style iterative algorithm. This context quantization strategy is applied to an embedded wavelet coding scheme in which the significance map symbols and sign symbols are directly coded by arithmetic coding with context models designed by the proposed quantization algorithm. Good coding performance is achieved.