International Journal of Pattern Recognition and Artificial Intelligence

In the present paper, we assess the performance of information-theoretic inspired risk functionals in multilayer perceptrons with reference to the two most popular ones, Mean Square Error and Cross-Entropy. The information-theoretic inspired risks, recently proposed: HS and HR2 are, respectively, the Shannon and quadratic Rényi entropies of the error; ZED is a risk reflecting the error density at zero errors; EXP is a generalized exponential risk, able to mimic a wide variety of risk functionals, including the information-thoeretic ones. The experiments were carried out with multilayer perceptrons on 35 public real-world datasets. All experiments were performed according to the same protocol. The statistical tests applied to the experimental results showed that the ubiquitous mean square error was the less interesting risk functional to be used by multilayer perceptrons. Namely, mean square error never achieved a significantly better classification performance than competing risks. Cross-entropy and EXP were the risks found by several tests to be significantly better than their competitors. Counts of significantly better and worse risks have also shown the usefulness of HS and HR2 for some datasets.

Time series shapelets are subsequences of time series that could be representative of a class. Shapelets-based time series classification methods can be divided into two large categories. The first category integrates shapelets selection within the process of constructing classifier; while the second category disconnects the process of finding shapelets from the classification algorithm by adopting a shapelet transformation. However, there are two important limitations of shapelet transformation. First, the number of shapelets selected for transformation has great influence on classification result, but it is difficult to decide the quantity of shapelets which yields the best data for classification. Second, similar shapelets always exist among the selected shapelets in previous algorithms. In our work, the latter problem is addressed by introducing an efficient and effective pruning technique, it filters similar shapelets and decreases the number of candidate shapelets at the same time. Then, we propose a novel shapelet coverage method to select shapelets for a given dataset. The final selected shapelets are named after Discriminative Shapelets. Our experimental results demonstrate that, on the classic benchmark datasets used for time series classification, shapelet pruning and coverage method outperforms ShapeletFilter.

In order to overcome the drawbacks of the K-means (KM) for clustering problems such as excessively depending on the initial guess values and easily getting into local optimum, a clustering algorithm of invasive weed optimization (IWO) and KM based on the cloud model has been proposed in the paper. The so-called cloud model IWO (CMIWO) is adopted to direct the search of KM algorithm to ensure that the population has a definite evolution direction in the iterative process, thus improving the performance of CMIWO K-means (CMIWOKM) algorithm in terms of convergence speed, computing precision and algorithm robustness. The experimental results show that the proposed algorithm has such advantages as higher accuracy, faster constringency, and stronger stability.

Locally linear embedding (LLE) depends on the Euclidean distance (ED) to select the k-nearest neighbors. However, the ED may not reflect the actual geometry structure of data, which may lead to the selection of ineffective neighbors. The aim of our work is to make full use of the local spectral angle (LSA) to find proper neighbors for dimensionality reduction (DR) and classification of hyperspectral remote sensing data. At first, we propose an improved LLE method, called local spectral angle LLE (LSA-LLE), for DR. It uses the ED of data to obtain large-scale neighbors, then utilizes the spectral angle to get the exact neighbors in the large-scale neighbors. Furthermore, a local spectral angle-based nearest neighbor classifier (LSANN) has been proposed for classification. Experiments on two hyperspectral image data sets demonstrate the effectiveness of the presented methods.

In this paper, we address a new scheme for symbol retrieval based on bag-of-relations (BoRs) which are computed between extracted visual primitives (e.g. circle and corner). Our features consist of pairwise spatial relations from all possible combinations of individual visual primitives. The key characteristic of the overall process is to use topological relation information indexed in BoRs and use this for recognition. As a consequence, directional relation matching takes place only with those candidates having similar topological configurations. A comprehensive study is made by using several different well-known datasets such as GREC, FRESH and SESYD, and includes a comparison with state-of-the-art descriptors. Experiments provide interesting results on symbol spotting and other user-friendly symbol retrieval applications.

We present a self-directed method for image segmentation using a modified top-down region dividing (TDRD) approach. The TDRD-based image segmentation method solves some of the issues with histogram and region growing-based segmentation techniques. The process is efficient and achieves proper results without over segmentation or spatial-structure destruction. In this paper, we examine seven user-defined parameters of the method. These parameters are converted from human inputs to values derived from in-class information created by the algorithm allowing for autonomous image segmentation, without the need of human input or feedback. Our new autonomous implementation also reduces the computational complexity of the algorithm. This reduction will produce significant savings for the total number of computations the algorithm needs to perform image segmentation. Experimental results show that the images using these new derived values yield superior results as compared to other methods, including the original TDRD method. We compare our results visually and numerically based on the within-class standard deviation (WCSD) and the number of connected components (NCC).

This paper proposes the Multilevel Gaussian Mixture Model (ML_GMM), a novel algorithm for online, unsupervised modeling of complex motion trajectories which allows for fast and reliable abnormal trajectory detection in video surveillance. The presented approach does not directly deal with target tracking or attribution of an observation to either one or another trajectories belonging to different targets. Instead, it receives as an input a sequence of observations and considers them as produced by a reliable, yet possibly noisy, tracker. As an output, it then produces a labeling of the input sequence, as a "normal" or "abnormal" whole trajectory. The normal and abnormal trajectory domains are learned in an unsupervised fashion during the training phase and can be updated while the classifier is running, thanks to its reduced computational complexity. The novelty of the contribution is twofold. First, ML_GMM is capable of representing short-term, as well as long-term, transitions in a consistent way, thus allowing for accurate representation of long and complex trajectories and their reliable labeling as normal or abnormal. Second, it allows the learning process to be run in parallel with the discrimination process, in order to support adaptability of the model to evolving operating conditions. Experimental results show that ML_GMM outperforms a number of comparable approaches in the literature.

Facial expression recognition has been researched much in recent years because of their applications in intelligent communication systems. Many methods have been developed based on extracting Local Binary Pattern (LBP) features associating different classifying techniques in order to get more and more better effects of facial expression recognition. In this work, we propose a novel method for recognizing facial expressions based on Local Binary Pattern features and Support Vector Machine with two effective improvements. First is the preprocessing step and second is the method of dividing face images into nonoverlap square regions for extracting LBP features. The method was experimented on three typical kinds of database: small (213 images), medium (2040 images) and large (5130 images). Experimental results show the effectiveness of our method for obtaining remarkably better recognition rate in comparison with other methods.

The biometric community is faced with the difficult problem of protection of the original biometric template. One way of doing this is using a cancelable biometric method, which transforms original biometric template in a noninvertible way and uses the transformed template to verify a person's identity. In this paper, we propose two novel representation methods for fingerprint minutiae. Proposed methods based on this representation are simple to generate cancelable templates without requiring pre-alignment of the fingerprints. The main idea is to generate a minimal spanning tree (MST) for fingerprint minutiae in a three-dimensional (3D) feature space. The chain code representation for a generated MST in a two-dimensional (2D) feature space is proposed. A bit string is then generated by mapping the chain code into 2D array. The fingerprint minutiae based upon the Cartesian system is dealt with in Method-1, while the boundary representation of the minutiae is dealt with in Method-2. The proposed methods are evaluated using FVC2004 and FVC2002 databases and the performance is better compared to existing methods [C. Lee and J. Kim, J. Netw. Comput. Appl.33(3) (2010) 236–246; S. Wang and J. Hu, Pattern Recogn.45 (2012) 4129–4137; S. Wang and J. Hu, Pattern Recogn.47(3) (2014) 1321–1329; Z. Jin, A. B. J. Teoh, T. S. Ong and C. Tee, Expert Syst. Appl.39 (2012) 6157–6167; P. Das, K. Karthik and B. C. Garai, Pattern Recogn.45(9) (2012) 3373–3388; A. B. J. Teoh, D. C. L. Ngo and A. Goh, Pattern Recogn.37(11) (2004) 2245–2255.]

In the age of Web 2.0, generalizing concepts are mainly based on information sharing and user cooperation. The user-centric Internet mode encourages more users to participate in the Internet. However, this mode causes the flooding of social networks with low-quality information. Thus, identifying trusted information in online social networks (OSNs) so as to enable the Internet to serve humans better has gradually become a research hotspot. However, most studies evaluate the quality of user-created content or discriminate reliable users in isolation. These results are specific to particular social networks and lack generality. Intuitively, reliable content and users are closely related. In this study, the authors attempt to combine these two types of reliable data by separately identifying reliable posts and users in a social network. Posts and users are found to improve each other based on the preliminary identification results. To deal with imbalanced data, an algorithm that combines oversampling and undersampling is used to build balanced data. An ensemble classifier is adopted for data classification. Experiments show that the proposed framework is both effective and efficient for most types of OSNs. The contributions of this study are two-fold: (i) a framework combining user-created content and reliable user recognition is proposed, and (ii) an ensemble classifier is built for use in data classification.