Face Recognition In Outdoor Dynamic Scene

Face recognition has been a hot research and application direction in computer vision area, which plays an important role among video surveillance, automatic guard system, national defense security and biometric recognition. In recent years, the research direction on face recognition algorithms has gradually transferred from constrained situations to unconstrained complex situations. Compared to static scene, the outdoor dynamic reality scene brings more challenging factors to face recognition such as the complex illumination changes, face pose variations and changeable background. So it is very applicable for us to study robust face recognition algorithms in complex dynamic reality scene.Based on the features of the dynamic scene, this thesis has done research on face alignment, feature extraction and classification method.The main contents of this thesis are as follows:1. We propose our own face crop and normalization method, which shows promising result in dealing with the change of illumination and face pose in dynamic scene. Furthermore we studied the common face preprocessing algorithms including Histogram Equalization, Logarithmic transformation and Gamma Intensity correction.Then we mainly focus on the improved algorithms based on Active Shape Model(ASM).2.On the basis of full analysis with regard to traditional face feature extraction method, we propose a high dimensional feature extraction using the image pyramid multi-scale technique, which improves the ability to describe human face characteristics. Then we studied the influence on recognition rate between PCA and sparse-based dimension reduction technique, and implemented the final face recognition algorithm using Joint Baysian method as well as support vector machine. The comparative experiment shows that our face recognition method demonstrate superior to the traditional ones.3.We studied some face feature fusion based methods and compared the classic SVM-based method to the probabilistic SVM. In the end, we proposed a new multi-feature fusion algorithm based on the probabilistic SVM4.On the basis of face preprocessing and face recognition algorithms, we propose our face preprocessing strategy, human face detection strategy and recognition decision making strategy in the outdoor dynamic scene. At last, we designed and implemented a face recognition system in the outdoor dynamic reality scene. Further more, we proved the accuracy and robustness of the system by experiment.