Research Of Face Recognition Based On Kernel Bayesian Compressive Sensing

Face recognition technology in the field of biometrics is the most natural and intuitive, friendly means, public security, justice, e-commerce, security, security monitoring, fields are widely used, face recognition technology generally includes two procedures:face detection and face recognition. First, detection and location in video information such as position and size of the face. Then, by using databases to identify matching, thereby identifying the target face, environments such as illumination, facial gestures, occlusion and facial expression changes has brought great challenges to face recognition.Theory of compressed sensing is a hot spot, it breaks the traditional limits of the Nyquist theorem, data acquisition at the beginning of time, removing superfluous information information is compressed and then reconstruction algorithm can revert to the original data from the sparse coefficient. Later scholars have combined with Bayesian compressive sensing algorithms, signal reconstruction using Bayesian statistics. Through the posterior distribution functions to estimate the signal and noise, so when the signal is restored, can reduce noise interference to some extent.Bayesian compressive sensing algorithms for introduction of hyper-parameters, when solving the posterior probability distribution of time complexity. Bag models and local character of traditional statistics are feature extraction method, Word bag model uses k-means clustering to the Visual Dictionary, cluster size selection and updating the dictionary takes a very long time. Local characteristics of statistical feature extraction algorithm of a light and rotational invariance, but face many changes on the scale of the effect is not very ideal.To tackle the above problem, we propose three new algorithms:nuclear Bayesian compressive sensing algorithm, compressed sparse pyramid and characteristics of multilayer local statistics. Bayesian compressive sensing nuclear unclear extension is the BCS, accelerating the operation rate of Bayesian compressive sensing, improve the robustness of the algorithm. CSPM is the idea of using compressed sensing the word bag model is improved, improved operation efficiency. Characteristics of multilayer local statistics is the statistical and spatial pyramid model with local characters, the algorithm can be overcome to a certain extent image do not match the problems people face, and verify the validity of the algorithm on the Extended Yale b, AR, CMU PIE,FERET face database.