基于人脸特征的快速分形人脸编码

在分析了人脸固有的特征后,提出了快速分形人脸编码算法,包括按值域块方差大小来动态确定w值,将定义域块候选区确定在与值域块呈对称分布的区域,利用DCT系数的能量分布来确定各域块的方向,从而减少对称变换个数。对该文算法时间复杂性的分析表明,利用该文算法所需时间要低于基于方差的方法。在ORL和YALE人脸数据库上的实验表明,在基本保持恢复图象质量的同时,该文算法所需编码时间仅为基于方差方法的1/4,证明了该文算法的有效性,实验还给出了算法中各参数的合适取值。

Fast Face Fractal Encoding Based on Features of Human Face

After analyzing the inherent features of human face,the fast fraetal face encoding algorithm is proposed.It is included that the value of w is determined dynamically according to the variance of range block,the region distributed symmetrically with the domain block is considered as the candidate region for domain blocks,the direction of blocks is calculated by the energy distribution of the DCT coefficients which lead to reduce the number of symmetrical transformation.The analysis on the time complexity of the proposed algorithm shows that the time cost of the proposed algorithm is lower than that of the variance based method.The experimental results on the ORL and YALE face databases show that the encoding time using the proposed algorithm is only 1/4 of that of the variance based method with the quality maintenance of the reconstruction image,which proves the proposed algorithm is effective.The appropriate value of the parameters in the algorithm is also given by the experiments.