Image Set Oriented Feature Extraction And Classification Methods

Currently,it is definitely convenient to obtain multiple images for subjects in many real applications,image set classification has become a hot research point in pattern recognition and computer vision.Compared with the traditional single image classification task,image set based classification can dramatically tackle an extensive collection of appearance variations within images including: variations in illumination,viewpoint changes,and occlusions.However,there exist several difficulties to handle in this task.For example,representing an image set by single mode can not effectively exploit the latent discriminative feature,parametric methods typically suffer from the problem that the query image set has weak statistical correlations with the training sets,considering that the image set is located in a high-dimensional space and there are redundant features,effective feature extraction is of great importance for classification.From the perspectives of collaborative representation,manifold learning,semi-supervised learning,and linear regression reconstruction,this paper proposes a series of methods to deal with the image set classification problem.The main works can be described as follows:(1)We propose a novel theoretical framework of set oriented multiple kernel learning for dimensionality reduction based on collaborative representation classification.To achieve this framework,we integrate the learning of an optimal kernel from the multiple base kernels and a discriminative projection into a unified formulation.Moreover,robust feature information can be effectively extracted by minimizing the intra-class reconstruction residual and maximizing the inter-class reconstruction residual of the regularized hull modeled for the image sets.Since the criterion of feature extraction conforms to the mechanism of the collaborative representation classifier,the collaborative representation coefficients in our model can be much discriminative across classes.Extensive experiments on benchmark datasets well demonstrate the effectiveness of the proposed method.(2)We propose a semi-supervised fuzzy discriminative learning framework to facilitate more robust image set classification.By using the semi-supervised setting which definitely has access to the labeled training data and the available unlabeled testing data,we adopt manifold distance metric to construct a fully trusted graph structure and derive two new data dependent probabilistic kernels to strongly reflect the underlying connection relationships between the training and query Gaussian manifold components.The resulted kernel representations are eventually integrated into a kernel fuzzy discriminant framework to enhance the compactness of intra-class Gaussian components and enlarge the margin for inter-class Gaussian components.Thus,more discriminating power of our learning machine is obtained for the classification of the query image set.Extensive experiments on several datasets well demonstrate the effectiveness of the proposed method compared with other image set algorithms.(3)We propose a novel multi-model fusion method across Euclidean space to Riemannian manifold to jointly accomplish dimensionality reduction and metric learning.To achieve our framework,we first introduce three distance metric learning models to better exploit the complementary information of an image set.Then,we aim to simultaneously learn a mapping performing dimensionality reduction and a metric matrix by integrating the two heterogeneous Euclidean space and Riemannian manifold into the common induced Mahalanobis space in which within-class data sets are close and between-class data sets are separated.This strategy can effectively handle the severe drawback of not considering the distance metric learning when performing dimensionality reduction in the existing set based methods.Extensive experiments on face recognition,object classification,gesture and handwritten classification well demonstrate the effectiveness of the proposed method compared with other image set algorithms.(4)Based on the concept of dual linear regression classification method,we propose a novel discriminative framework to exploit the superiority of discriminant regression mechanism.We aim to learn a projection matrix to force the represented image points from the same class to be close and those from different class are better separated.The feature extraction strategy in our discriminative framework can appropriately work with the corresponding classification strategy,thus,better classification performance can be achieved.Moreover,we propose a kernel discriminative extension method to address the non-linearity problem by adopting the kernel trick.From the experimental results,our proposed method can obtain competitive recognition rates on face recognition tasks via mapping the original image sets into a more discriminative feature space.Besides,it also shows the effectiveness for object classification task with small image sizes and different number of frames.