Abstract – This report intends to discuss Gabor feature method and the coupled PCA and Gabor system method used for face recognition. PCA system has also been used for face recognition which projects original set of face images onto lower dimensional face space. A Gabor filter

is a band-pass filter whose impulse response is given by multiplying a Gaussian envelope function with a complex sinusoidal. Gabor filters are used through which the image is passed and histogram values of the output image are dealt as Gabor features for the classification by finding the minimum Euclidean distance between the test image and the input image. PCA and Gabor feature method are then combined and the results compared against the results obtained using PCA and Gabor feature method separately. 

INTRODUCTION

Face recognition is an important aspect due to its potential applications such as person identification, human-computer interactions, security systems and image retrieval systems. Over the past decades considerable amount of time has been spent in the field of face recognition and various algorithms have been developed and implemented. The most basic of all

the algorithms is the Principal Component Analysis algorithm that is based on extracting the most prominent features of the face represented by a set of eigenvectors which are the principal components of our training set. These characteristic feature images are called ’eigenfaces’ [2]. The recognition part includes the projection of the new image onto the subspace spanned by the eigenfaces and then the classification is done on the basis of the distance between the new and the already known images. Gabor features are extracted using Gabor filters. These features include image’s local and discriminating features. The vector set formed by using the Gabor output and plotting its histogram is used to represent the image. The minimum Euclidean distance is calculated between the vectors of the test image and the image in the training set for the classification of the image. Though PCA and Gabor systems work well with modified images but both the systems fail for few modified images. So both the systems are coupled to give better results.