Two-dimensional principal component analysis (2DPCA) is a subspace technique used for facial image representation and recognition. Standard 2DPCA may be unable to extract informative features to adequately describe the inherent structural information of the original facial images with the presence of irrelevant variations such as lighting conditions, facial expressions, and so on. To deal with this, an improved fractional two-dimensional principal component analysis (IF2DPCA) is proposed in this paper. It is an extension of fractional 2DPCA (F2DPCA), which was developed based on the concept of fractional covariance matrix (FCM). IF2DPCA employs the same principle as F2DPCA for learning a projective matrix but further extends the use of fractional transformed 2D images throughout the entire recognition task. As a result, the feature subspace modeled by IF2DPCA maintains the most informative content of the 2D face images and is relatively insensitive to irrelevant variations. Experimental results on three face datasets confirm the effectiveness of the suggested IF2DPCA method in facial recognition.
Key words: Face Recognition, Feature Extraction, Fractional Covariance Matrix, 2DPCA, F2DPCA.
|
