Jun Kong

Generalized l P-regularized representation for visual tracking

Abstract Collaborative representation has been successfully applied to visual tracking to powerfully use all the PCA basis vectors in the target subspace for object representation. However, collaborative representation always exists redundant features that may affect the performance of visual tracking. In this paper, a visual tracking algorithm is proposed by solving a generalized l p -regularized (0≤p≤1) problem within a Bayesian inference framework for the reduction of redundant features. To efficiently solve the minimization problem of l p -regularization, the Generalization of Soft-threshold (GST) operator is applied in the framework of iterative Accelerated Proximal Gradient (APG) approach. Moreover, the GST operator can also provide a unified framework to observe the effects of different sparsity for visual tracking. To show the feasibility of l p -regularizer, we choose the representative l 0.5 -norm as the regularizer for the target coefficient and adjust the corresponding sparsity to be appropriate. Furthermore, we also introduce an extra l 0 -regularized tracker to observe the effect of excessive sparsity in a unified framework. Experimental results on several challenging sequences demonstrate that the proposed tracker leads to a more favorable performance in terms of accuracy measures including the overlap ratio and center location error, respectively.

GB(2D)2 PCA-based convolutional network for face recognition

Abstract. Face recognition is a challenging task in computer vision. Numerous efforts have been made to design low-level hand-crafted features for face recognition. Low-level hand-crafted features highly depend on prior knowledge, which is difficult to obtain without learning new domain knowledge. Recently, ConvNets have generated great attention for their ability of feature learning and achieved state-of-the-art results on many computer vision tasks. However, typical ConvNets are trained by a gradient descent method in supervised mode, which results in high computational complexity. To solve this problem, an efficient unsupervised deep learning network is proposed for face recognition in this paper, which combines both 2-D Gabor filters and (2D)2 PCA to learn the multistage convolutional filters. To speed up the calculation, the learned high-dimensional features are further encoded using short binary hashes. Finally, the obtained output features are trained using LinearSVM. Extensive experimental results on several facial benchmark databases show that the proposed network can obtain competitive performance and robust distortion-tolerance for face recognition.