基于最小二乘支持向量机误差校正的多方向插值算法

视频在易出错的信道中传输会发生数据丢包和误码等现象,严重影响视频图像质量。为了改善出错图像的质量,首先使用错误隐藏技术中的空域多方向插值方法进行插值,然后利用最小二乘支持向量机学习插值像素点之间的误差相关性来预测插值误差,进而对使用多方向插值方法得到的宏块进行误差校正。实验结果表明,与目前采用的最近像素线性插值法和多方向插值法相比较,基于误差校正的多方向插值算法在错误隐藏视觉效果和峰值信噪比性能指标上都具有一定的优越性。

Least Squares Support Vector Machines Error Correction Scheme Based Multi directional Interpolation

Compressed video streams are very vulnerable to transmission errors. Channel errors, including bit errors and packet loss, not only corrupt the current frame, but also propagate to subsequent frames. A novel multi directional interpolation scheme based on least squares support vector machines (LSSVM) error correction is proposed to get more satisfying error concealment results. At first, a multi directional interpolation for spatial error concealment is used to restore the missing block. Then LSSVM is trained with the interpolation error distribution of missing block is obtained from large local neighborhood of surrounding pixels of the source image. At last, the more accurate result frame is obtained by adding the estimated error distribution to the restored frame. Experimental results show that the visual quality and performance of the proposed method is improved compared to both nearest pixel linear interpolation and multi directional interpolation.