• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Arabian Journal for Science and Engineering >Feature-Learning Shrinkage Prediction for Spatial Error Concealment
【24h】

Feature-Learning Shrinkage Prediction for Spatial Error Concealment

机译：空间误差掩盖的特征学习收缩预测

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

For improving the overall performance of spatial error concealment (SEC), this paper proposes a feature-learning shrinkage prediction (FLSP) algorithm so as to obtain a better complexity-quality trade-off. Based on a basic concealment unit with sixteen-pixel neighbor and eight prediction directions, the FLSP algorithm innovatively adopts the non-iterative reconstruction mechanism with a shrinkage filling order. The local feature information of each missing block is firstly learned through a gradient detector, and then a multi-directional predictor is used to quickly recover the spatially correlated feature information of an extrapolation region. In each missing block, different pixel groups are group-by-group concealed according to their neighbor-level availabilities, and different pixels in a pixel group are one-by-one concealed according to a prior filling rule. By the experiment-driven methodology, the optional gradient detectors and prior filling rules are also analyzed for the FLSP algorithm. Compared with other SEC algorithms, experimental results show that the proposed FLSP algorithm can achieve better overall reconstruction performance under various loss conditions and thus strike a competitive trade-off between the computational complexity and reconstruction quality.
机译：为了提高空间错误隐藏（SEC）的整体性能，本文提出了一种特征学习收缩预测（FLSP）算法，以获得更好的复杂度-质量权衡。 FLSP算法基于具有16个像素邻域和8个预测方向的基本隐藏单元，创新地采用了具有收缩填充顺序的非迭代重建机制。首先通过梯度检测器学习每个丢失块的局部特征信息，然后使用多方向预测器快速恢复外推区域的空间相关特征信息。在每个丢失的块中，根据不同像素组的邻居级别可用性对它们进行逐组隐藏，并且根据先前的填充规则对像素组中的不同像素进行逐个隐藏。通过实验驱动的方法，还针对FLSP算法分析了可选的梯度检测器和先前的填充规则。与其他SEC算法相比，实验结果表明，所提出的FLSP算法在各种损失条件下都能获得较好的整体重建性能，从而在计算复杂度和重建质量之间取得了竞争取舍。

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