提出了改进的基于子空间投影的残缺人脸图像识别新算法，利用非负稀疏矩阵将原始人脸图像分解，从高维空间降到低维子空间，再利用Fisher线性判别式方法进一步降低子空间维数；然后利用欧式距离测量法，识别待测试人脸。实验证明，对残缺人脸图像的识别，新算法具有满意的识别率。An improved subspace method is proposed for recognizing imprecisely fragmentary faces.This method integrates non-negative matrix factorization with sparseness constraints and fisher’s linear discriminant analysis subspace. Comparative experiments show that our novel subspace has recognition performance better than PCA subspace in recognizing imprecisely fragmentary faces.