By Topic

Pattern Recognition, 2009. CCPR 2009. Chinese Conference on

Date 4-6 Nov. 2009

Filter Results

Displaying Results 1 - 25 of 209
  • 2009 CCPR - Title page

    Page(s): i
    Save to Project icon | Request Permissions | PDF file iconPDF (88 KB)  
    Freely Available from IEEE
  • 2009 CCPR [Copyright notice]

    Page(s): i
    Save to Project icon | Request Permissions | PDF file iconPDF (139 KB)  
    Freely Available from IEEE
  • 2009 CCPR Sponsors and Organization

    Page(s): i - iii
    Save to Project icon | Request Permissions | PDF file iconPDF (222 KB)  
    Freely Available from IEEE
  • 2009 CCPR Forewords

    Page(s): i - iii
    Save to Project icon | Request Permissions | PDF file iconPDF (213 KB)  
    Freely Available from IEEE
  • 2009 CCPR Keynotes

    Page(s): i - iv
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (239 KB)  

    Minimally supervised machine learning methods based on bootstrapping are an attractive approach to advanced information extraction. Complex patterns signalling relevant semantic relations in free texts can be detected in this way. However, the potential and limitations of such methods are not yet sufficiently understood. We have systematically analyzed a bootstrapping approach. The starting point of the analysis is a pattern-learning graph, which is a subgraph of the bipartite graph representing all connections between linguistic patterns and relation instances exhibited by the data. It is shown that the performance of such general learning framework for actual tasks is dependent on certain properties of the data and on the seed construction. However, the greatest improvements can be obtained through the systematic learning of negative patterns. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • 2009 CCPR Table of contents

    Page(s): i - xxvi
    Save to Project icon | Request Permissions | PDF file iconPDF (344 KB)  
    Freely Available from IEEE
  • 2009 CCPR author index

    Page(s): i - iii
    Save to Project icon | Request Permissions | PDF file iconPDF (100 KB)  
    Freely Available from IEEE
  • 3D Gray Level Moment Invariants: A Novel Shape Representation

    Page(s): 1 - 4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (397 KB) |  | HTML iconHTML  

    3D moment invariants are traditionally based on region characteristics and the location of every pixel point, this will cause high calculation cost. In this paper, a novel shape representation named 3D gray level moment invariants is constructed. Some properties of the new representation including the independence of the translation, scaling and rotation transforms are proved. Experiments indicate an encouraging high recognition rates without reducing the recognition performance compared with traditional methods. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Character Detection Algorithm in DCT Domain for Video

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (530 KB)  

    A DCT-domain based character detection algorithm is proposed for video stream. It utilizes the directional features of the texture in character blocks and the property that the characters in video usually distribute in row or column. The character/non-character blocks are effectively separated by a new adaptive threshold, then the noise and false text regions are further removed by morphological operation. Finally, the text regions are accurately obtained by horizontal and vertical projection. Experimental results demonstrate that the proposed approach can detect characters accurately even in video with complex background , and it is of good robustness and high practical value. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Classification Approach to Identify Definitions in Aviation Domain

    Page(s): 1 - 7
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (378 KB)  

    In this paper, we introduce a classification approach to identify definitions of all terms from a aviation professional corpus. The corpora of aviation domain are firstly segmented by LTP platform from HIT. Then four feature selection methods and two classifiers are applied to extract definitions. First of all, we summarize the correct proportion of feature subset used in classification of term definitions, and secondly argue that the naive Bayes classifier combined with CHI or ODDS for feature selection achieve the best score in the Fl-measure and F2-measure. In the end, we recognize that the use of SVM classifier with linear kernel could achieve very high precision, but the worst recall. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Classifying and Exploring System Based on Users' Tagging Behaviors

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1054 KB)  

    Nowadays, as information explosion, it becomes increasingly important for users to find a resource fast and efficiently in social tagging systems. To deal with the problem, this paper constructs an information classifying and exploring system based on users' tagging behaviors. We group the tags and resources by their semantic relations to construct Tag Bundles automatically, and generate a suitable category name for each group according to our category knowledge database, which generated by the open Web information resources. Meanwhile, the system allows users to browse their interests by the histogram and analyze their interests' changing. Finally, we build a prototype system to validate the effectiveness of the proposed method. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Design of Iris Recognition System at a Distance

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (342 KB) |  | HTML iconHTML  

    Iris recognition is a powerful biometrics for personal identification, but it is difficult to acquire good-quality iris images in real time. For making iris recognition more convenient to use, we design an iris recognition system at a distance about 3 meters. There are many key issues to design such a system, including iris image acquisition, human-machine-interface and image processing. In this paper, we respectively introduce how we deal with these problems and accomplish the engineering design. Experiments show that our system is convenient to use at the distance of 3 meters and the recognition rate is not worse than the state-of-the-art close-range systems. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Discretization Algorithm of Continuous Attributes Based on Supervised Clustering

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (330 KB) |  | HTML iconHTML  

    Many machine learning algorithms can be applied only to data described by categorical attributes. So discretization of continuous attributes is one of the important steps in preprocessing of extracting knowledge. Traditional discretization algorithms based on clustering need a pre-determined clustering number k, also typically are applied in an unsupervised learning framework. This paper describes such an algorithm, called SX-means (Supervised X-means), which is a new algorithm of supervised discretization of continuous attributes on clustering .The algorithm modifies clusters with knowledge of the class distribution dynamically. And this procedure can not stop until the proper k is found. For the number of clusters k is not pre-determined by the user and class distribution is applied, the random of result is decreased greatly. Experimental evaluation of several discretization algorithms on six artificial data sets show that the proposed algorithm is more efficient and can generate a better discretization schema. Comparing the output of C4.5, resulting tree is smaller, less classification rules, and high accuracy of classification. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Fast 2D Otsu Thresholding Algorithm Based on Improved Histogram

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (295 KB) |  | HTML iconHTML  

    Otsu adaptive thresholding is widely used in classic image segmentation. Two-dimensional Otsu thresholding algorithm is regarded as an effective improvement of the original Otsu method. To reduce the high computational complexity of 2D Otsu method, a fast algorithm is proposed based on improved histogram. Two-dimensional histogram is projected onto the diagonal, which forms 1D histogram with obvious peak and valley distribution. Then two-dimensional Otsu method is applied on a line that is vertical to the diagonal to find the optimal threshold. Furthermore, three look-up tables are utilized to improve the computational speed by eliminating the redundant computation in original two-dimensional Otsu method. Theoretical analysis and experimental simulation show that the proposed approach greatly enhances the speed of thresholding and has better immunity to salt and pepper noise. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Fast Rich Information-Based Stereo Matching Framework

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1348 KB) |  | HTML iconHTML  

    With the recent development on image affine region descriptors, we can extract more salient and useful local information from images. That information can be used to help us to better solve a fundamental problem in computer vision stereo vision. In this paper we propose a framework for stereo matching problems in order to give a rich-information based, high-precision and fast solution. Affine regions based SIFT are chosen as features rather than point features to extract more information. In the matching period, a search algorithm with incremental dissimilarity approximations is used for efficient computing. For correctness, MLESAC (maximum likelihood estimation sample consensus) method is used to eliminate outliers. In the experiment part, we evaluate different combinations on the performance of speed, correctness and transformations. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Fault Diagnosis Method Combined Fuzzy Logic with CMAC Neural Network for Power Transformers

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (273 KB) |  | HTML iconHTML  

    Dissolved gas analysis (DGA) is an effective method for early detection of incipient faults in power transformers. To improve the accuracy of fault diagnosis, a fault diagnosis method combined fuzzy logic with cerebellar model articulation controller (CMAC) neural network is proposed in this paper. The proposed fuzzy CMAC neural network (FCMAC) has an optimization mechanism to ensure high diagnosis accuracy for all general fault types. Firstly, it uses fuzzy logic to extract diagnosis rules from a lot of fault samples, and then, the extracted rules are employed to optimize CMAC network. Many real fault samples are analyzed by FCMAC for the purpose of verification, and the analyzed results are also compared with those analyzed by IEC ratio method and those by the CMAC neural network. The comparison results show that the proposed method has remarkable diagnosis accuracy. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Fuzzy Clustering Algorithm for Image Segmentation Using Dependable Neighbor Pixels

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (455 KB) |  | HTML iconHTML  

    In this paper, a fuzzy clustering algorithm using dependable neighbor pixels is proposed for image segmentation. In order to enhance the segmentation performance, the proposed algorithm utilizes the local statistical information to discriminate dependable neighbor pixels from undependable neighbor pixels, and then allows the labeling of the pixel to be influenced by the dependable neighbor pixels. This algorithm has two advantages: (1) the spatial information with high reliability is incorporated into the objective function so that the segmentation accuracy is guaranteed; (2) the intensity of the spatial constraints is automatically determined by the similarity meature so that the segmentation result is adaptive to the original image. The efficiency of the proposed algorithm is demonstrated by extensive segmentation experiments using both synthetic and real images. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Graph Representation for Silhouette Based on Multiscale Analysis

    Page(s): 1 - 3
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (246 KB) |  | HTML iconHTML  

    Graph descriptor is usually focused on in computer vision for its flexibility and richness. However, in object recognition, it is difficult to catch the feature of an object completely with a straightforward way by a graph. In this paper, from a multiscale viewpoint, we propose a method to construct a vertex-labeled graph for image recognition where the label represents the importance of a vertex to the graph. By using the Fourier Descriptor, when we adjust the cut-off frequency we can select how much detailed feature can be used to represent a contour. In this process, we found the medial axis of shape also evolves from a simple graph to a detailed graph. Focusing on this evolution, for each vertex in the graph we assign a label representing its importance in this graph. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Method Based on General Model and Rough Set for Audio Classification

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (382 KB)  

    As one of important information component in multimedia, audio enriches information perception and acquisition. Analyses and extractions of audio features are the base of audio classification. It's important to extract audio features effectively for content-based audio retrieval. In this paper, based on the theory of rough set, audio features are reduced and a lower-dimension feature set can be obtained with more effective. Then the feature set is applied in the general model for audio classification. Experiments show that this method is effective. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Method of Building Chinese Basic Semantic Lexicon Based on Word Similarity

    Page(s): 1 - 4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (238 KB) |  | HTML iconHTML  

    Identification of sentiment orientation in Chinese words is essential for getting sentiment comprehension of Chinese text, and building a basic semantic lexicon with Chinese emotional words will provide a core subset for identifying emotional words in a special area. It can not only help to identify and enlarge semantic lexicon in corpus effectively but also improve classification efficiency. On the basis of the similarity of Chinese words, the paper has proposed a method of calculating sentiment weight of Chinese emotional words. In addition, a dictionary with basic Chinese emotional words has been constructed based on the HowNet semantic lexicon. By utilizing the dictionary together with TF-IDF, we have done experiments to identify sentiment orientation in Chinese text and have got satisfying classification result. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Modified Differential Evolution Algorithm for Multi-Objective Optimization Problems

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (402 KB)  

    Differential evolutionary (DE) is a simple, fast and robust evolutionary algorithm for multi-objective optimization problems (MOPs). This paper is to introduce a modified differential evolutionary algorithm (MDE) to solve MOPs. There are some different points between MDE and traditional DE: individual mutation and its selection strategy; MDE allows infeasible solutions of population to participate in mutation process, and mutation strategy of individuals adapt to a modified updating scheme of particle velocity in PSO. The fast nondominated sorting and ranking selection scheme of NSGA-II proposed by Deb is incorporated into individual's selection process. We finally obtain a set of global optimal solutions (gbest). Simulated experiments show that the obtained solutions present good uniformity of diversity, and they are close to the true frontier of Pareto. Also, the convergence of solutions obtained is satisfactory. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A Multi-Class Multi-Manifold Learning Algorithm Based on ISOMAP

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (311 KB) |  | HTML iconHTML  

    The classical algorithm ISOMAP can find the intrinsic low-dimensional structures hidden in high-dimensional data uniformly distributed on or around a single manifold, but if the data are sampled from multi-class, each of which corresponds to an independent manifold, and clusters formed by data points belonging to each class are separated away, several disconnected neighborhood graphs will form, which leads to the failure of ISOMAP algorithm. In this paper, an improved version of ISOMAP, namely multi-class multi-manifold ISOMAP (MCMM-ISOMAP), is proposed. MCMM-ISOMAP constructs a single neighborhood graph not by increasing the value of neighborhood parameter, but by the following steps that first choose appropriate value with which short-circuit edges can not be introduced, second find such pairwise data each of which are two endpoints of the shortest Euclidean distance between classes, and finally make them neighborhood points each other. Thereby a single neighborhood graph will form, and then ISOMAP algorithm is applied to find the intrinsic low-dimensional embedding structure. Experimental results on synthetic and real data reveal effectiveness of the proposed method. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A New Approach for Video Scene Boundary Detection

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (543 KB)  

    Scene is the semantic unit in video. Video scene segmentation is a difficult task in content based video structure analysis. This paper proposes a new approach for scene boundary detection. We first construct shot content coherence signal using normalized cut criterion and then use a heuristic algorithm to detect scene boundary. Because the normalized cut criterion simultaneously emphasizes on the inhomogeneity of shots in different scenes and the homogeneity of shots in the same scene, the continuous signal reflects the coherence of shot content well. Experiments on different kinds of video clips demonstrate our approach performs well in scene segmentation. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A New Binarization Method for a Sign Board Image with the Blanket Method

    Page(s): 1 - 4
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (595 KB) |  | HTML iconHTML  

    We propose a new binarization method suited for character extraction from a sign board in a scenery image. The binarization is one of a significant step in character extraction in order to get high quality result. Character region on sigh board, however, has many variation and colors. In addition to it, if there exists high frequency texture region like a mountain or trees in the background, it can be a cause of difficulty to binarize an image. At the high frequency region, the binarized result is sensitive to the threshold value. On the other hand, a character region of sign board consists of solid area, that is, includes few high frequency regions, and has relatively high contrast. So the binarized result of character region is stable at an interval of the threshold value. Focusing attention on this point, we propose a new binarization method which obtains a threshold value based on the fractal dimension by evaluating both region's density and stability to threshold value. Through the proposed method, we can get a fine quality binarized images, where the characters can be extracted correctly. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A New Mixed Particle Filter Based on an Auxiliary Model

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (300 KB)  

    Particle filter is an effective method for non-linear filter and it has been gained special attention of researchers in various fields. There will be a new mixed particle filter (PUPF) proposed in this paper based on the general particle filter and the unscented particle filter. lt first uses the general particle filter to generate particles for estimating the state at time k and then a new auxiliary model will be introduced. We would use the unscented particle filter to estimate the state at time k the second time. This structure makes use of the latest observation information, it has small error and better stability. The experimental results indicate that the proposed particle filter's performance outperforms the other four particle filters .The result indicates that the PUPF is a useful method for nonlinear filter problems. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.