任何一种单子空间特征提取算法都不能在任何情况下优于其他子空间算法,但是采用双子空间却可以克服单子空间的局限性。为了提高分类结果的正确率,提出了一种基于PCA及ICA的双空间特征提取算法,该算法采用ICA作为PCA的补空间进行特征提取，其目的是将在PCA子空间中难以识别的样本，再次投影到ICA子空间中进行识别。该算法可分为以下两个步骤:首先进行预分类，即在一个子空间内同时使用两种分类器对测试样本进行分类,若某个测试样本被两种分类器划分到不同的类，则将该测试样本加入到新测试样本集中；然后将新测试样本集中的测试样本再次投影到另一个子空间中进行分类识别；最后，将识别结果与预分类结果一起进行正确率测试。在ORL及FERET人脸库上的实验结果表明，该算法的模式识别率明显优于传统的特征提取算法。

No single subspace feature extraction algorithm outperforms others under all circumstances, but applying dual space feature extraction algorithm can overcome the limits of single subspace. To increase the recognition rate of classified results, a new dual space feature extraction algorithm is put forward based on PCA (principal component analysis) and ICA (independent component analysis). The algorithm uses ICA as PCA’s complementary space to realize feature extraction. The aim of the algorithm is to reproject the samples into ICA subspace to classify, which are difficult to be classified in PCA subspace. The proposed dual space extraction algorithm includes two steps. Step 1 is called preclassify, which uses two classifiers in a single subspace to classify the testing samples. If a testing sample is classified into different classes by two classifiers, it would be added into the new testing samples set. In step 2, the algorithm projects the samples of the new testing samples set into another subspace and classifies them. Then, the recognition results of two subspaces are tested together. The experimental results in ORL and FERET database prove the dual space feature extraction algorithm outperforms the traditional feature extraction algorithm in recognition rate.