根据人脸识别中光照方向类别多，类与类之间特征不明显，不易分类的问题，提出了采用SVM分类方法解决人脸识别中复杂的、非线性光照方向分类问题。用分叉树将复杂分类变成每级分类，并保证在4类以内的分类方法；同时在各级分类中，根据当前分类的类别特点，提取当前要处理的分类之间的明显特征作为特征进行分类；另外，给出了最优的SVM分类器的训练和构造过程。实验结果表明，该方法对解决分类特征不易于提取，类别数目多的分类问题有明显效果，分类准确率达到89．16％。

To solve the complex and nonlinear illumination direction classification problem in face recognition, a method based on bifurcate tree and SVM is proposed in this paper. Before using the SVM, light directions are firstly sorted into a bifurcate tree according to its intrinsic intensity distribution which assure that only a few classes are sorted at every branch. Moreover, different features are extracted for SVM at the different levels according to the lighting characteristics of the corresponding levels. Experiment results show that the proposed method is efficient to solve multi-classes classification problem and the accuracy can be achieved to 89.16%.