本文运用深度神经网络的方法克服姿态变量和图像分辨率的影响,提出了一种多姿态的人脸超分辨识别算法并在实验数据集上获得了良好的性能表现。另外本文利用深度信念网络探索正面人脸图像和侧面人脸图像的映射,方法放松了深度信念网络的输入也输出之间绝对相等,而只是保证其高层含义上的相等。实验表明了使用深度信念网络可以学习到侧面人脸图像到正面人脸图像的一个全局映射,但是丢失了个体细节差异。本文还提出了基于深度网络保持姿态邻域进行姿态分类的方法,在学习过程中,我们保证了同一个姿态下的人脸图像应该与同一姿态下的多张图像互为邻居。实验证明了,我们的方法在用于姿态分类具有非常良好的性能,但是也发现学习过程中,那些与区别个体的信息逐渐丢失了,这也导致了直接运用非线性近邻元分析的特征的人脸识别的性能不佳。-In this paper, the neural network approach to overcome the depth of variables that affect the attitude and image resolution , proposed a multi-pose face recognition algorithms and super-resolution experimental data set obtained in a good performance. Also this paper to explore the depth of belief network mapping frontal face image and profile face images , the method of absolute equality between the input relax depth of belief networks is also output , but only to ensure equal meaning on its top . Experimental results show that the use of deep belief networks can learn to face image to the side of the front face of a global image map , but lost the details of individual differences . This paper also proposes to maintain posture neighborhood depth network-based gesture classification methods in the learning process , we ensure that the face image under the same gesture with multiple images should be under the same attitude are neighbors . Experiment proves that our method for gesture cl