基于支持向量机的图象插值及错误隐匿策略

如何对在有损网络环境中传输的视频进行错误隐匿是视频传输研究中的基本问题。支持向量机（SVM）是一种新兴的通用学习算法，是国际上机器学习领域新的热点。为了取得比现有方法更好的错误隐匿效果，提出了一种新的基于支持向量机回归估计的错误隐匿策略，首先建立了基于支持向量机回归估计的图像插值算法，并将其引入到错误隐匿问题中，然后用空域插值的方法达到错误隐匿的目的。实验结果表明，与目前采用的各种错误隐匿策略相比较，基于支持向量机的错误隐匿策略在错误隐匿效果和推广性能上都具有一定的优越性。

Image Interpolation and Error Concealment Scheme Based on Support Vector Machine

How to prevent quality degradation due to channel errors for images and video transmitting over lossy channels is a fundamental problem in multimedia signal processing. Support Vector Machine(SVM) is a novel powerful learning method and is now a new hotspot in the field of machine learning, and has been successfully used in many pattern recognition problems. To get more satisfying error concealment results, a novel error concealment scheme based on SVM is proposed in this paper. At first, a SVM based image interpolation algorithm is successfully established. SVM learning machines are carefully trained by a large amount of training data exacted from standard images to learn the relationship of neighboring pixels in spatial domain, and theses well trained machines are used as specific nonlinear interpolation operators. Comparative results show that this kind of interpolation operator outperforms not only some traditional used interpolation operators such as linear and median operators, but also some operators carefully trained by artificial neural networks. The error concealment problem is placed into a spatial image interpolation framework and the proposed interpolation method is thoroughly used to estimate the missing image blocks according to their neighboring pixels. Experimental results show that compared with some error concealment schemes both in spatial and frequent domain in the literature, especially those based on artificial neural networks, the proposed one has remarkable advantages in error concealment performance and generalization property.