Comparative analysis of 3D face recognition using 2D-PCA and 2D-LDA approaches

Even if, most of 2D face recognition approaches reached recognition rate more than 90% in controlled environment, current days face recognition systems degrade their performance in case of uncontrolled environment which includes pose variations, illumination variations, expression variations and ageing effect etc. Inclusion of 3D face analysis gives an age over 2D face recognition as they give vital informations such as 3D shape, texture and depth which improve discrimination power of an algorithm. In this paper, we have investigated different 3D face recognition approaches that are robust to changes in facial expressions and illumination variations. 2D-PCA and 2D-LDA approaches have been extended to 3D face recognition because they can directly work on 2D depth image matrices rather than 1D vectors without need for transformations before feature extraction. In turn, this reduces storage space and time required for computations. 2D depth image is extracted from 3D face model and nose region from depth mapped image has been detected as a reference point for cropping stage to convert model into a standard size. Two Dimensional Principal Component Analysis (2D-PCA) and Two Dimensional Linear Discriminant analysis (2D-LDA) are employed to obtain feature vectors globally compared to feature vectors obtained locally using PCA or LDA. Finally, euclidean distance classifier is applied for comparison of extracted features. A set of experiments on GavabDB 3D face database, which has 61 individuals in total, demonstrated that 3D face recognition using 2D-LDA method has achieved recognition accuracy of 93.3% and EER of 8.96% over database, which is higher compared to 2D-PCA. So, more optimized performance has been achieved using 2D-LDA for 3D face recognition analysis.