Abstract: In unconstrained condition, illumination, posture and facial expression becomes the main choke point of wild face recognition. Based on this issue, a fusion algorithm is proposed with learning prototype hyperplanes and side-Information based linear discriminant analysis is proposed. Support Vector Machine（SVM） model is used to each sample in weak labeled data set to be mid-level feature of prototype hyperplanes, and SVM sparse set is selected from generic data set without labeled by learning combination coefficient. Fisher discriminative criterion is used to maximize discriminat ability under the constraint of combination sparse coefficient of SVM model, and the objective function is solved by iterative optimization algorithm. SILD is used to extract features and cosine similarity measure is used to finish face recognition. The effectiveness and reliability of proposed algorithm has been verified by experiments on the two common databases Extended YaleB and labeled face of wild（LFW）. The results show that proposed algorithm has better recognition efficiency comparing with several other face recognition algorithms.

Key words: wild face recognition, mid-level feature representation, Support Vector Machine, Singular Value Decomposition, side-information based on linear discriminant

摘要： 非限制环境下光照、姿势、表情等变化已成为户外人脸识别的主要瓶颈所在。针对这一问题，提出了一种学习原型超平面融合线性判别边信息的算法进行人脸识别。利用支持向量机将弱标记数据集中的每个样本表示成一个原型超平面中层特征；使用学习组合系数从未标记的通用数据集中选择支持向量稀疏集；借助于Fisher线性判别准则最大化未标记数据集的判别能力，并使用迭代优化算法求解目标函数；利用线性判别边信息进行特征提取、余弦相似性度量以完成最终的人脸识别。在Extended YaleB和户外标记人脸（LFW）和通用人脸数据集上进行实验，验证了所提算法的有效性和可靠性。实验结果表明，相比其他几种较为先进的人脸识别算法，所提算法取得更好的识别性能。

关键词: 户外人脸识别, 中层特征表示, 支持向量机, 奇异值分解, 线性判别边信息