The Design Of Face Recognition System

Biometric technology is tightly integrated computer science,physics,ecology and biostatistics to build models and do identification with human inherent physiological and behavioral characteristics such as fingerprints,face image,iris image,voice,gait, etc.Face recognition is an important area of biometric technology,it analysis and comparison face image and feature information to do biological identification with computer.Face recognition is also called portrait recognition and it automatically detect the location of the human face and related technical processing from the acquired images.With the development of science and technology the demand of face recognition technology is increasing and it has received a great attention and improvement.In this paper,the course of action of face recognition system design is,firstly,using the camera to captured image and do pre-processing operations to uniform the original including image filtering and image enhancement.Secondly,using the face detection mothod based on shape feature technology to detecte and locate face positions in the image.Atlast,extract facial features to do the matching and recognition job.In reality because the different of distance between the camera and individual,the size of face and calibration scale will be diffrent and it make influence on the recognition effect.In order to solve this problem a new mothod is proposed in this paper and it is combining discrete cosine transform and principal component analysis combination. In general,after the image discrete cosine transform the image energy is concentrated in the upper left corner of the discrete cosine transform domain image.To retain the upper left corner of the low frequency information and appropriately discard high frequency information enable different scales face images having the same DCT coefficients so the scales of face images become the same.Extract the same dimension DCT coefficient matrix that shows the primitive face image, thus making the problem of original image dimensions resolved. In the actual study it is generally desirable to use a few variables to represent the original data in most of the variables, and these variables is the main component of the image,using principal component analysis to reduce dimension, the use of a certain number of principal component scores instead of the original data so the characteristic dimension of the image can be decreased. Then use the modulus normalized method for image matching recognition and the main idea of it is using the geometric meaning of the angle between the vectors to determine the cosine similarity between images. The correctness of the method is verified by experiment,in the experiment the principal components is fixed at first,to verify the impact of the discrete cosine transform coefficients recognition rate by change the discrete cosine transform coefficients.Then discrete cosine transform coefficients is fixed and to verify that the primary effect of a fraction of the recognition rate by change the principal components,and finally the least significant parameter estimates can be calculated.