For the high time complexity problem of Markov random field(MRF)model in face recognition with pose variation, this paper proposed a multi-scale MRF algorithm framework.This algorithm coupled super coupled transform (SCT)to optimi-zing the child problems during the MRF processing,which had successfully reduced the complexity of MRF.Firstly,it fin-ished symmetrical image match by two images.Then,it used multi-scale local binary patterns (MLBP)histogram feature to generate distance ratio.Finally,it got the final ratio comparison and paired each pair of zones.Experimental results on labeled face database in wild (LFW),the XM2VTS database and the CMU-PIE database show that both recognition rate and speed of proposed algorithm is superior to other excellent algorithms.Performance on GPU framework basically meets the needs of prac-tical application.%针对马尔可夫随机场模型（MRF）在姿态变化人脸识别中复杂度较高的问题，提出了一种多尺度 MRF算法，通过串联超级耦合变换（SCT）优化多尺度 MRF 过程的子问题，明显降低了 MRF 的计算复杂度。首先，通过两层图像完成对称匹配；然后，利用多尺度局部二值模式（MLBP）直方图特征生成距离比；最后，比较获得最终比值，配对每对区域。在户外标记人脸数据库（LFW）、XM2VTS 数据库和 CMU-PIE 数据库上进行实验，实验结果表明，相比现有的优秀算法，该算法在识别率和执行速度方面均表现优越，在图形处理器上基本满足实际应用需求。
展开▼
