Abstract: An image compression algorithm based on the dependencies between the Contourlet coefficients is proposed. Through the mutual information computation, it is discovered that strong dependencies exist in local intra-band micro-neighborhoods, and that the shape of these neighborhoods is highly anisotropic in the different decomposition bands. So the local intra-band micro-neighborhoods are used to predict the  current coefficient and the multi-objective restraint optimization method is introduced to obtain the predictor weight value. The prediction precision is enhanced effectively by the method. This algorithm is exploited in a Contourlet-based image coding application and it is showed that the decoded image has better detail and the textural property than JPEG2000 under certain compression ratio situation.

Key words: image compression, Contourlet transform, dependency in intra-band, prediction method, multi-objective restraint optimization method

摘要： 提出一种基于Contourlet变换域子带内系数相关性的图像压缩算法。通过互信息量的计算，发现Contourlet变换域子带内紧邻的系数之间存在较强的相关性，并且在不同的分解子带中系数之间的相关性强弱呈现出位置上的各向异性。在此基础上，利用局部邻居信息预测当前系数，同时引入多目标约束优化方法得到预测器的权重值，可有效提高预测精度。通过与JPEG2000算法进行实验对比，在一定的压缩比情况下，采用该算法解码后的图像具有较好的细节和纹理特征。

关键词: 图像压缩, Contourlet变换, 子带内相关性, 预测算法, 多目标约束优化