Face Recognition Algorithms Based On Image Reconstruction And Features Fusion

The visual information of faces play a very important role in human communication. Face recognition technique tries to make the computer have the ability of identifying person by analyzing and extracting features from the face images. Because of its wide applications, the face recognition has been one of the hot research subjects in near four decades.A mature face recognition process can be divided into two steps: feature representation and classification. The extraction of image features is one of the fundamental tasks in image recognition. From features extraction angle of view, we proposed some novel methods for face recognition. The computer simulations illustrate these methods enhance the recognition rate effectively and have theoretical and practical values. The main contributions of this thesis include:1) Face recognition based on image reconstruction: Traditional subspace-based Face Recognition methods obtain a universal subspace by using all trained images. The subspace mainly represents the commonness of human faces but there are a few sights of the individuality owned by a single person's face. In this paper, based on subspace methods and image reconstruction, we present some novel methods for face recognition. When applied to face recognition, the fundamental difference between the traditional subspace methods and our methods is that we obtain the basis images by using each person's pictures respectively, while the traditional way uses the whole training images of the database. After the step above, we obtained the features which would be employed to reconstruct the images by mapping the test images to the basis images. And then we use two ways for face recognition, the first way is adopting the minimum reconstruction error and the second is employing support vector machine (SVM) by using the reconstruction error vectors. Finally, experiments based on three different databases illustrate the effectivity of these methods.2)Face Recognition Based on Features Fusion: Support vector machine (SVM) is a new machine learning method that is established on the statistics learning theories, and in this paper, we propose a novel algorithm for facial recognition based on features fusion in support vector machine (SVM). First, some local features and global features from pre-processed face images are obtained. The global features are obtained by making use of discrete cosine transform (DCT) and singular value decomposition (SVD). At the same time, the local features by utilizing non-negative matrix factorization (NMF) are also obtained. Furthermore, the feature vectors fused by independent component analysis (ICA) with global and local features are given. Finally, the feature vectors are used to train SVM to realize the face recognition, and the computer simulation illustrates the effectivity of this method on the ORL face database.3) Independent Features Fusion for Face Recognition: As a holistic feature extraction method, the DCT converts high-dimensional face images into low-dimensional spaces in which more significant facial features are maintained. On the contrary, the Gabor transformed face images exhibit strong characteristics of spatial locality, scale and orientation selectivity, which produce salient local features which are most suitable for face recognition. So, we use Gabor wavelets for the local features and then integrate Gabor features with DCT coefficients. In addition to, because ICA would reduce redundant features and represent more explicitly the independent features which are most useful for subsequent pattern discrimination and associative recall, ICA has been used for extracting their independent features respectively. And then, the independence property of the independent Gabor features and DCT features leads to the application of the support vector machine (SVM) for classification.4) Face Recognition Based on HMM-SVM: Because SVM has excellent ability to classify and HMM has good ability to time sequence modeling, we used a mixed model based on HMM and SVM for face recognizing. In this method, a sequence of overlapping sub-images is extracted from each face image by using discrete cosine transform (DCT) and singular value decomposition (SVD). Afterward, the sequence which is extracted from training images is modeled by using HMM, and then, the output probability of each HMM for the training sequence has been considered as the input vector of SVM for its training. In the end, the output probability of each HMM for the testing sequences has been considered as the input vector of SVM for its testing. The computer simulation illustrates the effectivity of this method on the Olivetti research laboratory (ORL) database.5) Face recognition prototype system design: Based on these novel algorithms, we discuss the process of face recognition and introduce the function and implementation modules, and then develope the face recognition prototype system. All these results lay a foundation for a business version in the future.