Abstract
Linear discriminant analysis (LDA) is one of the most popular subspace techniques widely used in face recognition (FR) to extract low dimensional features that discriminate a set of facial images belong to a certain classes. In LDA-based FR, 2D images modeled as 1D image vectors. The high dimensionality and singularity problem are two important drawbacks of such modeling. An alternative version of LDA has been proposed to resolve these drawbacks named two-dimensional linear discriminant analysis (2D-LDA) which directly operates on 2D facial image matrices. 2D-LDA significantly lowered the computational cost of LDA, and straightforward in features extraction, hence, possess high discriminatively and accuracy, as well as it completely alleviated singularity problem. Further, subspace-based FR methods have been implemented in wavelet domain motivated by the fact that mapping images from pixels space to that of wavelets can provide enhanced features carrying the most informative information of the images in reduced dimensionality. Likewise other 2D subspace methods, 2D-LDA only reflects the correlation between image rows when evaluating the scatter matrices. An improvement has been made to these methods via correlating the information between image rows and that of columns by using diagonal facial images. Accordance to aforementioned observations, a FR method is proposed in this paper in which diagonal 2D-LDA performed in wavelet subspace aiming to implicitly combine the beneficial properties of 2D-LDA, wavelet transform (WT) and diagonal image in one framework to achieve improved recognition performance. The competitively of proposed method is demonstrated by experimental results on standard face dataset.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have