Pattern Recognition: Methods and Applications

Pattern recognition - Methods and Applications includes contributions from university educators and active research experts. This book is intended to serve as a basic reference on pattern recognition, especially on the topics related to image and graphics processing, shape analysis, text processing, and bioinformatics analysis.

Chapter 1 proposes a review of traditional outlier detection methods and their recent enhancements. Some particular data representations are presented. A case-study based on a synthetic data is proposed in order to demonstrate the potential of a fuzzy logic approach which combines several techniques.

Chapter 2 studies the conditions for which the solution given by the Maximum Entropy Principle is equivalent to that given by Support Vector Machines. It describes a unifying framework that computes the probability density function and the optimal separating surface from examples.

Chapter 3 presents techniques for building multi-sensor fusion classifiers. A pairwise diversity-based ranking strategy is introduced to select a subset of ensemble components, which when combined, will be more diverse than any other component subset of the same size.

Chapter 4 proposes a neural-network-based differential evolution approach for face recognition (FR). The approach combines neural network classifiers and differential evolution updates, applies both 2D texture and 3D surface feature vectors, and effectively enhance the FR performance.

Chapter 5 proposes an efficient margin-based linear embedding method that exploits the nearest hit and the nearest miss samples only.

Chapter 6 proposes an efficient algorithm to construct the skeleton of a binary image as a geometric graph whose edges are Bezier curves 1 and 2 degrees.

Chapter 7 presents a novel structured light means with no coding procedure involved. By projecting a binary rhombic pattern, and computing the 3D surface normal at grid-points, the 3D reconstruction procedure can be realized via the proposed surface integration methods.

Chapter 8 focuses on exploring the potential of virtual worlds in order to train appearance-based models for pedestrian detections in ADAS. A de facto pedestrian detector is used for this task: a linear SVM with HOG features.

Chapter 9 proposes a technique for partially automating the creation of a large-scale dictionary or corpus. More specifically, this involves adding unknown words to an existing language resource; in this case, a thesaurus.

Chapter 10 proposes a probabilistic model that explicitly considers the document relations represented by links. A given document is modeled as a mixture of a set of topic distributions, each of which is borrowed (cited) from a document that is related to the given document.

Chapter 11 proposes a improve of the segmentation method by topic ClustSeg. The proposed improvement is a strategy to automatically calculate the threshold for deciding the cohesiveness between textual units. This proposal can be used by other methods of text segmentation by topic.

Chapter 12 proposes a comparative analysis of Wavelets, used as input attributes of support vector machines, which will be responsible for classification of pathological voices.

Chapter 13 introduces a complete methodology for automatic human chromosome classification. The methodology isolates the chromosomes from microscopic images, extracts their characteristic band profiles, and then classifies them.

Chapter 14 presents a compositional spectra approach for classifying bacterial genomes. The problem of bacteria classification arose long before the start of the Genomic Era.

Chapter 15 describes how spectroscopic and chromatographic methods coupled to pattern recognition multivariate algorithms can be an excellent tool for the determination of fuel compliance to technical specifications and origin determination purposes.

Pattern recognition - Methods and Applications includes contributions from university educators and active research experts. This book is intended to serve as a basic reference on pattern recognition, especially on the topics related to image and graphics processing, shape analysis, text processing, and bioinformatics analysis.

Chapter 1 proposes a review of traditional outlier detection methods and their recent enhancements. Some particular data representations are presented. A case-study based on a synthetic data is proposed in order to demonstrate the potential of a fuzzy logic approach which combines several techniques.

Chapter 2 studies the conditions for which the solution given by the Maximum Entropy Principle is equivalent to that given by Support Vector Machines. It describes a unifying framework that computes the probability density function and the optimal separating surface from examples.

Chapter 3 presents techniques for building multi-sensor fusion classifiers. A pairwise diversity-based ranking strategy is introduced to select a subset of ensemble components, which when combined, will be more diverse than any other component subset of the same size.

Chapter 4 proposes a neural-network-based differential evolution approach for face recognition (FR). The approach combines neural network classifiers and differential evolution updates, applies both 2D texture and 3D surface feature vectors, and effectively enhance the FR performance.

Chapter 5 proposes an efficient margin-based linear embedding method that exploits the nearest hit and the nearest miss samples only.

Chapter 6 proposes an efficient algorithm to construct the skeleton of a binary image as a geometric graph whose edges are Bezier curves 1 and 2 degrees.

Chapter 7 presents a novel structured light means with no coding procedure involved. By projecting a binary rhombic pattern, and computing the 3D surface normal at grid-points, the 3D reconstruction procedure can be realized via the proposed surface integration methods.

Chapter 8 focuses on exploring the potential of virtual worlds in order to train appearance-based models for pedestrian detections in ADAS. A de facto pedestrian detector is used for this task: a linear SVM with HOG features.

Chapter 9 proposes a technique for partially automating the creation of a large-scale dictionary or corpus. More specifically, this involves adding unknown words to an existing language resource; in this case, a thesaurus.

Chapter 10 proposes a probabilistic model that explicitly considers the document relations represented by links. A given document is modeled as a mixture of a set of topic distributions, each of which is borrowed (cited) from a document that is related to the given document.

Chapter 11 proposes a improve of the segmentation method by topic ClustSeg. The proposed improvement is a strategy to automatically calculate the threshold for deciding the cohesiveness between textual units. This proposal can be used by other methods of text segmentation by topic.

Chapter 12 proposes a comparative analysis of Wavelets, used as input attributes of support vector machines, which will be responsible for classification of pathological voices.

Chapter 13 introduces a complete methodology for automatic human chromosome classification. The methodology isolates the chromosomes from microscopic images, extracts their characteristic band profiles, and then classifies them.

Chapter 14 presents a compositional spectra approach for classifying bacterial genomes. The problem of bacteria classification arose long before the start of the Genomic Era.

Chapter 15 describes how spectroscopic and chromatographic methods coupled to pattern recognition multivariate algorithms can be an excellent tool for the determination of fuel compliance to technical specifications and origin determination purposes.