International Journal of Pattern Recognition and Artificial Intelligence

Despite the relatively high accuracy of the naïve Bayes (NB) classifier, there may be several instances where it is not optimal, i.e. does not have the same classification performance as the Bayes classifier utilizing the joint distribution of the examined attributes. However, the Bayes classifier can be computationally intractable due to its required knowledge of the joint distribution. Therefore, we introduce a "pairwise naïve" Bayes (PNB) classifier that incorporates all pairwise relationships among the examined attributes, but does not require specification of the joint distribution. In this paper, we first describe the necessary and sufficient conditions under which the PNB classifier is optimal. We then discuss sufficient conditions for which the PNB classifier, and not NB, is optimal for normal attributes. Through simulation and actual studies, we evaluate the performance of our proposed classifier relative to the Bayes and NB classifiers, along with the HNB, AODE, LBR and TAN classifiers, using normal density and empirical estimation methods. Our applications show that the PNB classifier using normal density estimation yields the highest accuracy for data sets containing continuous attributes. We conclude that it offers a useful compromise between the Bayes and NB classifiers.

Most of the initialization approaches are dedicated to the partitional clustering algorithms which process categorical or numerical data only. However, in real-world applications, data objects with both numeric and categorical features are ubiquitous. The coexistence of both categorical and numerical attributes make the initialization methods designed for single-type data inapplicable to mixed-type data. Furthermore, to the best of our knowledge, in the existing partitional clustering algorithms designed for mixed-type data, the initial cluster centers are determined randomly. In this paper, we propose a novel initialization method for mixed data clustering. In the proposed method, both the distance and density are exploited together to determine initial cluster centers. The performance of the proposed method is demonstrated by a series of experiments on three real-world datasets in comparison with that of traditional initialization methods.

The subgraph isomorphism problem is one of the most important problems for pattern recognition in graphs. Its applications are found in many different disciplines, including chemistry, medicine, and social network analysis. Because of the -completeness of the problem, the existing exact algorithms exhibit an exponential worst-case running time. In this paper, we propose several improvements to the well-known Ullmann's algorithm for the problem. The improvements lower the time consumption as well as the space requirements of the algorithm. We experimentally demonstrate the efficiency of our improvement by comparing it to another set of improvements called FocusSearch, as well as other state-of-the-art algorithms, namely VF2 and LAD.

In a learning process, features play a fundamental role. In this paper, we propose a Boosting-based feature selection algorithm called BoostFS. It extends AdaBoost which is designed for classification problems to feature selection. BoostFS maintains a distribution over training samples which is initialized from the uniform distribution. In each iteration, a decision stump is trained under the sample distribution and then the sample distribution is adjusted so that it is orthogonal to the classification results of all the generated stumps. Because a decision stump can also be regarded as one selected feature, BoostFS is capable to select a subset of features that are irrelevant to each other as much as possible. Experimental results on synthetic datasets, five UCI datasets and a real malware detection dataset all show that the features selected by BoostFS help to improve learning algorithms in classification problems, especially when the original feature set contains redundant features.

In this paper, we consider the problem of image restoration with box-constraints. Image restoration problem is ill-conditioned and the regularization approach has widely been used to stabilize the solution. The restored image highly depends on the choice of the regularization parameter. The regularization parameter is generally determined by trial-and-error method when no true original image is available. Obviously, it is time consuming. The main aim in this paper is to develop an algorithm to choose the regularization parameter automatically when the box-constraints are imposed. In the proposed algorithm, the regularization parameter is adaptively determined by the previous iterative solution. Numerical simulations are used to demonstrate the performance of the proposed method.

Localization of texts in natural images could be an important stage in many applications such as content-based image retrieval, visual impairment assistance systems, automatic robot navigation in urban environments and tourist assistance systems. However due to the variations of font, script, scale, orientations, color, shadow and lighting conditions, robust scene text localization is still a challenging task. In this paper, we propose a novel method to localize not only Farsi/Arabic and Latin texts with different sizes, fonts and orientations but also low luminance contrast and poor quality ones in the natural images taken with uneven illumination conditions. Firstly, fast weighted median filtering as a nonlinear edge-preserving smoothing filter and then color contrast preserving decolorization are exploited to make the text localization system more robust for low luminance contrast and poor quality texts. In order to extract the Farsi/Arabic and Latin scene texts and also filter the nontext ones, a unified framework is proposed incorporating the maximally stable extremal regions and a novel proposed region detector called Stable Width Stroke Regions which is based on closed boundary regions. Phase congruency and Laplacian operators are exploited to extract the closed boundary regions. Finally, to extract the single text lines, the Meanshift clustering and radon transform were used. Experimental results show that the proposed method localize low luminance contrast and low quality scene texts for both Farsi/Arabic and Latin scripts encouragingly.

In this paper, visual attention spreading is formulated as a nonlocal diffusion equation. Different from other diffusion-based methods, a nonlocal diffusion tensor is introduced to consider both the diffusion strength and the diffusion direction. With the help of diffusion tensor, along with the principle direction, the diffusion has been suppressed to preserve the dissimilarity between the foreground and background, while in other directions, the diffusion has been boosted to combine the similar regions and highlight the salient object as a whole. Through a two-stages diffusion, the final saliency maps are obtained. Extensive quantitative or visual comparisons are performed on three widely used benchmark datasets, i.e. MSRA-ASD, MSRA-B and PASCAL-1500 datasets. Experimental results demonstrate the superior performance of our method.

Manual selection of features from massive unstructured point cloud data is a very time-consuming task that requires a considerable amount of human intervention. This work is motivated by the need of fast and simple algorithm to obtain robust, stable and well-localized interest points that are used for subsequent processing in computer vision real-time applications. This paper presents an algorithm for detection of interest points in three-dimensional (3D) point cloud data by using a combined 3D Sobel–Harris operator. The proposed algorithm is compared with six state-of-the-art approaches used to identify the true feature points. Extensive experiments were carried out using synthetic benchmark and real datasets. The datasets were selected with different sizes, features and scales. The results were evaluated against human generated ground truth and predefined feature points. Three measures were used to evaluate the algorithm accuracy, namely localization accuracy L_e, False Positive Error (FPE) and False Negative Errors (FNE). Also, the complexity analysis of the proposed algorithm is presented. The results show that the proposed algorithm can identify the interest points with accepted accuracy. It works directly on point cloud datasets and shows superiority when compared with other methods work on 3D mesh data.

This paper proposes a novel video-based vehicle detection approach with data-driven adaptive neuro-fuzzy networks. The key ideas include configuring several virtual loops as vehicle detection zones in the image, assuming moving vehicles will cause pixel intensities and local textures to change, and then identifying such changes to detect vehicles. In this work, vehicle detection is treated as a pattern classification problem. First, 14 image features (regarding foreground area, texture change, and environmental condition) are extracted to represent the distinction between vehicle and nonvehicle patterns. Then, three neuro-fuzzy networks are trained via incremental semi-supervised learning to build a data-driven adaptive classifier, which judges whether a vehicle is located in the virtual loop. The semi-supervised learning procedure is performed based on a modified tri-training approach, to automatically optimize the structures and parameters of the component neuro-fuzzy networks. Experimental results illustrate that the proposed approach is accurate and robust to detect vehicles in complex environments (e.g. adverse illumination and weather conditions), and thus can improve the performance of video-based vehicle detection.

In this paper, a new recognition algorithm was developed based on the skin color model and the fingertip structure detection to improve the hand gesture recognition accuracy based on the traditional Hu's moment features. Firstly, the geometric features and the areas of skin were adopted to segment the skin region from the background. The Douglas–Peucker (D–P) algorithm was utilized to conduct the contour approximation to get a polygonal in the process of feature detection. Then the convexity point inspections were conducted in the polygonal. Secondly, we developed two rules for locating and numbering the fingertips. After that, we built a seven-dimensional feature vector. Finally, the hand gesture was recognized by using the distance matching criterion. The developed recognition algorithm improved the recognition accuracy by 2.7% compared with the traditional Hu's moment features. Additionally, it has good robustness in several manners such as shifting, plane rotation and scaling.

Facial expressions can be mainly conveyed by only a few discriminative facial regions of interest. In this paper, we study the discriminative regions for facial expression recognition from video sequences. The goal of our method is to explore and make use of the discriminative regions for different facial expressions. For this purpose, we propose a Hidden Markov Model (HMM) Decision Forest (HMMDF). In this framework, each tree node is a discriminative classifier, which is constructed by combining weighted HMMs. Motivated by a psychological theory of "elimination by aspects", several HMMs on each node are modeled respectively for facial regions, which have discriminative capabilities for classification. The weights for these HMMs can be further adjusted according to the contributions of facial regions. Extensive experiments validate the effectiveness of discriminative regions on facial expression, and the experimental results show that the proposed HMMDF framework yields dramatic improvements in facial expression recognition compared to existing methods.

In the belief function theory, several measures of uncertainty have been introduced. One of their possible use is unreliable source discounting before the fusion stage. Two different measures of uncertainty exist which are the intrinsic and extrinsic ones. The intrinsic measure makes it possible to assess the source's confusion whereas the extrinsic one measures the contradiction between sources. In this paper, we associate both measures in order to estimate the global reliability of a source. This method, named Generic Discounting Approach (GDA), is proposed in two different versions: Weighted GDA and Exponent GDA. Those reliability measures are integrated into a classifier. The method was tested, against to some pioneer approaches, on several UCI datasets as well as on an urban image classification problem and showed very encouraging results.