3D Reconstruction and Face Recognition Using Kernel-Based ICA and Neural Networks

Abstract

Kernel-based nonlinear feature extraction and classification algorithms are popular research topics in machine learning. In this paper, we propose an improved photometric stereo scheme based on the basic reflectance model. In order to reconstruct a human face as a 3D model, we use kernel independent component analysis (KICA) to obtain the face’s surface normal vector on each point of the image. In this procedure, we find that the x-axis, y-axis and z-axis values of the normal vector’s coordinates are not arranged in order. Thus, an improved KICA (IKICA) method is proposed that takes the normal vector of a synthetic spherical surface normal vector as the supervised reference for solving this problem. After obtaining the correct normal vector’s sequence form surface, we use a method for enforcing integrability to reconstruct 3D objects. We test our algorithm on synthetically generated images to reconstruct object surfaces on a number of real images captured from the Yale Face Database B, and use three kinds of methods to fetch characteristic values. Those methods are called contour-based, circle-based, and feature-based methods. Then, a three layer feed-forward neural network trained by back-propagation algorithm is used to realize a classifier. All the experimental results were compared to those of the existing human face reconstruction and recognition approaches tested on the same images. The experimental results demonstrate that the proposed improved kernel independent component analysis (IKICA) method of reconstruction and human recognition are efficient approaches.