The Research Of Dimensionality Reduction Methods Based On Sparse In Face Recognition

Biometric technology is a technology which use the human characteristics to authenticate.It’s secure and reliable, and is not easy to forge, and can not be easily stolen and so on. Thefacial feature authenticate is the most natural and most direct means in all biometricidentification technology. In the face recognition technology,the characteristics of the humanface is extracted through the computer and then calculate.Face recognition technology has thecharacteristics of non-intrusive, simple operation and good concealment. Based on the aboveadvantages a growing number of scholars and organizations take a lot attention,so that theface recognition technology is one of the hottest research topics in the past30years in patternrecognition and image processing. And it has been widely applied in the field of identification,e-commerce, video surveillance, human-computer interaction, security and management,criminal reconnaissance.Although the human through the eyes can remember the thousands of different face,faster and easier to identify the different face, but let the computer automatically facerecognition is still a challenging.The identification of non-rigid objects are often moredifficult than the rigid object identification, a person’s face is very different due to agechanging, rich facial expressions, the shooting face imaging distances, the point of view of theimaging and different light, and many other factors. Brought great uncertainty to thecalculation of face recognition algorithms. In short, the face recognition contains multipledisciplines such as computer vision, image processing, neural networks, a very promising andchallenging technology.This paper first introduces the significance of face recognition and face recognitionresearch status. Then introduced the sparse representation, and, intuitive graphical vividlydescribes the sparse representation of related concepts. Leads to a linear regression model,in-depth discussion of the method of least squares, ridge regression, lasso regression, elasticnetwork and their advantages and disadvantages, and gives each implementation steps.Original face image can not directly reflect the nature of the object, due to the lightconditions, the shooting angle, the facial expressions, posture. So how to effectively extractthe face effective features is the most basic question of face recognition. Given a strong noisepollution or narrow face image, we still be able to easily recognize it out, which indicates thatwe do not need to perceive all the pixels in the image, as long as the perception of some ofwhich will be able to complete the identification task. This means that the human visualsystem has sparse image features. In this paper, a two-dimensional extension of sparsity preserving projections algorithm (2DSPP), the two-dimensional nature can reserves thespatial structure of the image information, although there is no discrimination information butthe sparse nature of the very natural retain the local characteristics of the image set, and noselection of parameters.Then we also proposed a new graph structure, each column of the image extracted fromseeking a close neighbor relationship, then the two images is close neighbor relationship tosee how many columns to each other in a close neighbor relationship, if the number ofcolumns by neighbors is larger than a given parameter values, we say that the two images inthe nearest neighbor relations, and this method is applied to two-dimensional localitypreserving projections algorithm (2DLPP) and experiment in ORL, YALE and AR, theexperimental results show that our method has good recognition rate than other methods.