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.