Multiclass Spectral Clustering Research Articles

Learning iterative quantization binary codes for face recognition

Binary feature descriptors have been widely used in computer vision field due to their excellent discriminative power and strong robustness, and local binary patterns (LBP) and its variations have proven that they are effective face descriptors. However, the forms of such binary feature descriptors are predefined in the hand-crafted way, which requires strong domain knowledge to design them. In this paper, we propose a simple and efficient iterative quantization binary codes (IQBC) feature learning method to learn a discriminative binary face descriptor in the data-driven way. Firstly, similar to traditional LBP method, we extract patch-wise pixel difference vectors (PDVs) by computing and concatenating the difference between center patch and its neighboring patches. Then, inspired by multi-class spectral clustering and the orthogonal Procrustes problem, which both are widely used in image retrieval field, we learn an optimized rotation to minimize the quantization error of mapping data to the vertices of a zero-centered binary hypercube by using iterative quantization scheme. In other words, we learn a feature mapping to project these pixel difference vectors into low-dimensional binary vectors. And our IQBC can be used with unsupervised data embedding method such as principle component analysis (PCA) and supervised data embedding method such as canonical correlation analysis (CCA), namely IQBC-PCA and IQBC-CCA. Lastly, we cluster and pool these projected binary codes into a histogram-based feature that describes the co-occurrence of binary codes. And we consider the histogram-based feature as our final feature representation for each face image. We investigate the performance of our IQBC-PCA and IQBC-CCA on FERET, CAS-PEAL-R1, LFW and PaSC databases. Extensive experimental results demonstrate that our IQBC descriptor outperforms other state-of-the-art face descriptors.

Blind image clustering based on the Normalized Cuts criterion for camera identification

Camera identification is a well known problem in image forensics, addressing the issue to identify the camera a digital image has been shot by. In this paper, we pose our attention to the task of clustering images, belonging to a heterogenous set, in groups coming from the same camera and of doing this in a blind manner; this means that side information neither about the sources nor, above all, about the number of expected clusters is requested. A novel methodology based on Normalized Cuts (NC) criterion is presented and evaluated in comparison with other state-of-the-art techniques, such as Multi-Class Spectral Clustering (MCSC) and Hierarchical Agglomerative Clustering (HAC). The proposed method well fits the problem of blind image clustering because it does not a priori require the knowledge of the amount of classes in which the dataset has to be divided but it needs only a stop threshold; such a threshold has been properly defined by means of a ROC curves approach by relying on the goodness of cluster aggregation. Several experimental tests have been carried out in different operative conditions and the proposed methodology globally presents superior performances in terms of clustering accuracy and robustness as well as a reduced computational burden.

3D Shape Partition via Multi-class Spectral Graph Clustering

This paper proposes a new 3D shape partition algorithm. The main idea of our method is to represent 3D shape as a weighted graph whose nodes denote each face in the shape and the weights on the edges encode the similarities between faces. We then extract a set of geometric features from shape to build an affinity matrix for weighted graph and use multi-class spectral clustering method to find an optimal partition. We compute the generalized eigenvalue problems of the affinity matrix to group the most similar faces into the same semantic part and apply the graph cut method to refine the boundaries between the partition parts. Finally we evaluate our approach on the Princeton shape benchmark. The experiment shows that our method can successfully segment different classes of 3D shapes and can achieve more meaningful segmentation results compared to other leading single shape partition algorithms.

Action recognition using linear dynamic systems

In this paper, we propose a novel approach based on Linear Dynamic Systems (LDSs) for action recognition. Our main contributions are two-fold. First, we introduce LDSs to action recognition. LDSs describe the dynamic texture which exhibits certain stationarity properties in time. They are adopted to model the spatiotemporal patches which are extracted from the video sequence, because the spatiotemporal patch is more analogous to a linear time invariant system than the video sequence. Notably, LDSs do not live in the Euclidean space. So we adopt the kernel principal angle to measure the similarity between LDSs, and then the multiclass spectral clustering is used to generate the codebook for the bag of features representation. Second, we propose a supervised codebook pruning method to preserve the discriminative visual words and suppress the noise in each action class. The visual words which maximize the inter-class distance and minimize the intra-class distance are selected for classification. Our approach yields the state-of-the-art performance on three benchmark datasets. Especially, the experiments on the challenging UCF Sports and Feature Films datasets demonstrate the effectiveness of the proposed approach in realistic complex scenarios.

Multi-Class Spectral Clustering Based on Particle Swarm Optimization

Spectral clustering has been used in computer vision successfully in recent years, which refers to the algorithm that the global optima is found in the relaxed continuous domain obtained by eigen decomposition, and then a multi-class clustering problem should be solved by traditional clustering algorithm such as k-means. In this paper, we propose a novel spectral clustering algorithm based on particle swarm optimization (PSO). The major contribution of this work is to combine PSO technique with spectral clustering. In the multi-class clustering stage, the PSO is applied in the feature space to cluster the new data, each of which is a characterization of the original data. Experimental studies on PSO-based spectral clustering algorithm demonstrate that the proposed algorithm provides global convergence, steady performance and better accuracy.