Human face recognition under pose variation with fusion geometric methods

Abstract

In recent years, different methods have investigated in order to extract effective and important features in face recognition. Apparent features include the coordinates of faceʼs comonents. Such as eyes, rings, tissues and different parts of the face and they are called apparent characteristics. Many limitations are imposed on the extraction of the features of the image. Main purpose of this research is to recognize humanʼs face using combining fusion methods under different situations. Each image can be considered as a matrix and algebra operations and mathematical transformations are performed on it. Image are two - dimentional and image points are considered statistical data and statistical characteristics of the points are used to describe image. In this method, those characteristics are used that have enough ability for image description, information richness Local Phase Quantization (LPQ) and divide face image into several parts using Local Binary Pattern (LBP) and describe image features such as illumination. Changes, perturbations and blur which are important in face recognition. We choose an image and perform LBP once and LPQ. Features extracted form image are considered as a vector using combining two above methods. If vector extracted from image have high dimension, they be reduced because classes can be separated a classified in this way better. After preparing image bank, features are extracted from people image of image bank and their dimensions are reduced. Then, image components become more clear and detailed for face description. We investigate how to appear face ofter combining histograms. Simulation experiments illustrate that combining two above methods has better recognition rate that single method