When it comes to the face recognition, this paper exceptionally focuses on the facial feature extraction based on subspace analysis. Firstly, this paper introduces the constitution of the face recognition system, and then analyses the key technologies, such as the face detection, feature extraction, and image pretreatment processing. It mainly analyses the various face recognition algorithms. According to the high efficiency of wavelet in the processing of image data matrix and the shortcoming of less dimension of the LDA training sample, PCA cannot use higher order statistical properties of the data. Combining these three algorithms, this paper puts forward the improved recognition method. Simulation experiments with MATLAB are carried out and the results show the effectiveness of the method.