Research Of Face Recognition Based On Algebraic Feature And DSP Implementation

Face recognition is currently the field of pattern recognition and machine vision research hot spots. The facial feature is a unique biological characteristics, with the convenience, immediacy, uniqueness, effectiveness, and many other ad- vantages, has been widely applied to national security, commercial confidentially, human-computer interaction, visual communication, information retrieval and so on. However, it is vulnerable to face the change of external environment, physiological, image conditions, and many other uncertain factors. They make face recognition a challenging research topic and fundamental problem.In recent years, the algebraic features of the feature extraction method for convenience of calculation, effective, and so favored by the majority of face recognition scholars, lots of scholars have proposed many improved algorithm used in face recognition. In this paper, we will discuss face recognition of the algebraic features. The work as follows.1. Facial images are pretreated and normalized to reduce the impact of background, lighting, hair, face expression and so on. Some important features are reserved for the face recognition. Introduce the concept of Assembled Matrix distance (AMD), it has been shown to face a better characterization of the degree of similarity between the image matrix.2. In this paper, based on the two-dimensional principal component analysis (2DPCA) , we proposed a feature extraction method based on SI2DPCA+AMD, which uses the symmetric properties of frontal face images, uses within-class scatter matrix instead of the total scatter matrix construct the covariance matrix, and combines with AMD metric distance to construct the face classifier to achieve the final classification. Experiments show that the method has good classification results.3. Describe the LDA as well as two-dimensional linear discriminant analysis(2DLDA), a method combines AMD metric distance and bidirectional and modular 2DLDA (BM2DLDA) which is proposed in the paper. First, the original images are divided into modular images, and then in rows and columns directions apply the 2DLDA, finally finished by AMD-based distance classifier classification. Experiments show that the method obtains a higher recognition rate. 4. Use the Matlab GUI platform design face recognition algorithm analysis platform, which can well show the performance of various face recognition algorithms. Besides, this chapter also introduces the TI TMS320DM6437 DSP of the Davinci series, and the face recognition codes were transplanted and optimized on the DSP.