• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Asia-Pacific Signal and Information Processing Association Annual Summit and Conference >Graph Learning Fast Transform Coding of 3D River Data
【24h】

Graph Learning Fast Transform Coding of 3D River Data

机译：3D River数据的图学习和快速转换编码

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Collection of physical river measurements across space and time is important towards analysis and prediction of dynamic river flow, and thus early warning and prevention of flood disaster. In this paper, we focus on lossy compression of 3D river data at high quality using predictive graph-based transforms. Specifically, we first divide 3-dimensional river data into non-overlapping temporal frame groups. Data in a frame group t is then predicted using frame group t-1, assuming strong temporal correlation. Then for each block in a frame in group t, we learn a sparse inverse covariance matrix from a spatial neighborhood of blocks in the previous frame group via a graphical lasso algorithm with structural constraints. The learned matrix is then interpreted as a graph Laplacian, and graph lifting transform (GLT) or fast graph Fourier transform (FGFT) are employed to encode the prediction residuals efficiently. Experimental results show coding performance gain over conventional DCT and competing graph transform schemes without graph learning.
机译：收集跨时空的实际河道测量数据对于分析和预测动态河道流量，从而对洪水灾害进行预警和预防非常重要。在本文中，我们专注于使用基于预测图的变换对3D河数据进行高质量的有损压缩。具体来说，我们首先将3维河流数据划分为不重叠的时间帧组。然后假设强时间相关性，然后使用帧组t-1预测帧组t中的数据。然后，对于组t中帧中的每个块，我们通过具有结构约束的图形套索算法从前一帧组中的块的空间邻域中学习稀疏逆协方差矩阵。然后将学习的矩阵解释为图拉普拉斯算子，并采用图提升变换（GLT）或快速图傅立叶变换（FGFT）来有效地编码预测残差。实验结果表明，在没有图学习的情况下，与传统的DCT和竞争图变换方案相比，编码性能有所提高。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号