自适应方向插值及相似性约束人脸区域的H.264视频流错误隐藏研究

Research of error concealment based on adaptive directional interpolation and facial region of similarity constrains for H.264 video stream

针对H.264视频压缩码流在信道传输过程中易发生丢包和误码错误,以及视频图像解码重建后人脸区域模糊等问题,提出了一种基于自适应方向插值及相似性约束人脸区域的错误隐藏算法,对错误块进行两次错误隐藏。首先利用自适应方向插值空域错误隐藏算法对错误块中每个像素点根据其纹理方向进行自适应方向插值运算并恢复,实现第一次错误隐藏;然后提取错误块视频图像的前景目标,采用肤色分割的方法定位出可能的人脸区域,再基于相似性约束方法确定用于覆盖错误块人脸区域的目标人脸区域,通过仿射变换后,将目标人脸区域的梯度信息覆盖错误块的人脸区域,实现第二次错误隐藏。改进的方法能够更加精确地确定插值方向且更好地恢复人脸区域,重建后的视频平滑性有一定的改善。采用JM 8.6平台仿真结果表明,与现有基于人脸五官特征的空域差错掩盖算法相比,视频图像隐藏效果在主观视觉上有一定改善,计算出的峰值信噪比(PSNR)提高了1.21 d B。

H. 264 compressed video stream occurred easily to data loss and bit errors in the transmission process, at the same time, face region in decoding reconstruction was fuzzy. Aiming at these questions, this paper proposed an error concealment algorithm based on adaptive directional interpolation and facial region of similarity constrains so as to carry on error concealment two times to error block. First, it used spatial error concealment algorithm based on adaptive directional interpolation to recover the error block by an interpolation to achieve the first error concealment. Then, it extracted the foreground object of error block to gain the target face region, using the facial partition method to locate the possible face region and calculating the similarity between adjacent frames and blocks. Last, after the affine transformation, it covered the face region of error block used gradient information of the target face region to achieve the second error concealment. The proposed method improve the accuracy of directional interpolation and the smoothness of restored video. JM 8.6 simulation experiments show that the proposed method outperforms spatial error concealment algorithm based on the facial features, with 1.21 dB gain in PSNR.