Creation of large scale face dataset using single training image

Face recognition (FR) has become one of the most successful applications of image analysis and understanding in computer vision. The learning-based model in FR is considered as one of the most favorable problem-solving methods to this issue, which leads to the requirement of large training data sets in order to achieve higher recognition accuracy. However, the availability of only a limited number of face images for training a FR system is always a common problem in practical applications. A new framework to create a face database from a single input image for training purposes is proposed in this dissertation research. The proposed method employs the integration of 3D Morphable Model (3DMM) and Differential Evolution (DE) algorithms. Benefitting from DE's successful performance, 3D face models can be created based on a single 2D image with respect to various illumination and pose contexts. An image deformation technique is also introduced to enhance the quality of synthesized images. The experimental results demonstrate that the proposed method is able to automatically create a virtual 3D face dataset from a single 2D image with high performance. Moreover the new dataset is capable of providing large number of face images equipped with abundant variations. The validation process shows that there is only an insignificant difference between the input image and the 2D face image projected by the 3D model. Research work is progressing to consider a nonlinear manifold learning methodology to embed the synthetically created dataset of an individual so that a test image of the person will be attracted to the respective manifold for accurate recognition.