论文部分内容阅读
目前可视电话等应用对人脸图像质量要求很高,为了解决这一问题,提出一种基于人脸五官特征的空域差错掩盖(FFSEC)算法。首先利用边缘检测技术对丢失块进行分类,以完成初步差错掩盖;然后使用肤色分割技术提取肤色区域;最后利用模式识别技术获取眼睛的具体位置,根据五官的相对位置以及具体特征对其进行插值差错掩盖。采用H.264的JM11模型对算法进行仿真验证,实验结果表明,FFSEC算法相对于现有的基于人脸特征的自适应差错掩盖算法(BFASEC),在丢包率较大的情况下,峰值信噪比(PSNR)提升了约0.5~1.5 dB,视觉效果提升明显。
Currently, applications such as video telephony are very demanding on face image quality. To solve this problem, a spatial error concealment (FFSEC) algorithm based on facial features is proposed. First of all, using the edge detection technology to classify the missing blocks to complete the initial error concealment; then using the skin color segmentation technology to extract the color area; Finally, using pattern recognition technology to obtain the specific location of the eye, according to the relative position of the features and specific features of its interpolation error cover. The algorithm of JM11 is used to simulate the algorithm. The experimental results show that, compared with the existing BFASEC based on face feature, the FFSEC algorithm can reduce the packet loss rate, The PSNR is improved by about 0.5 ~ 1.5 dB, and the visual effect is obviously improved.