Optimum selection of features for 2D (color) and 3D (depth) face recognition using modified PCA (2D)

The paper proposes a Modified Principal Component Analysis coined as 2DPCA to compare 2D and 3D face recognition. In 2DPCA a covariance matrix of image is obtained directly from the original image and is used to find the eigenvectors for image feature extraction. Here the Texas 3D [1] face recognition database was considered, which has 1149 pairs of high resolution, preprocessed and pose normalized color and range images. These images are pixel-to-pixel registered and of resolution of 751×501 pixels. The experiment performed using the images reconstructed from feature vectors demonstrated that depth information was beneficial in representing and recognizing the face with least number of principal components.