Pooja Saigal

Color image classification and retrieval through ternary decision structure based multi-category TWSVM

In this paper, we propose Ternary Decision Structure based multi-category twin support vector machines (TDS-TWSVM) classifier. Twin support vector machines (TWSVM) formulation deals with finding non-parallel plane classifier which is obtained by solving two related Quadratic Programming Problems (QPPs). The proposed TDS-TWSVM classifier is an extension of TWSVM so as to handle multi-class data and is more efficient in terms of training and testing time of classifiers. For a K-class problem, a balanced ternary structure requires Â? log 3 K Â? comparisons for evaluating a test sample. The experimental results depict that TDS-TWSVM outperforms One-Against-All TWSVM (OAA-TWSVM) and binary tree-based TWSVM (TB-TWSVM) in terms of classification accuracy. We have shown the efficacy of the proposed algorithm via image classification and further for image retrieval. Experiments are performed on a varied range of benchmark image databases with 5-fold cross validation.

Angle-based twin support vector machine

In this paper, a novel nonparallel hyperplane based classifier termed as “angle-based twin support vector machine” (ATWSVM) has been proposed which is motivated by the concept of twin support vector machine (TWSVM). TWSVM obtains two nonparallel hyperplanes by solving a pair of quadratic programming problems (QPPs). ATWSVM presents a generic classification model, where the first problem can be formulated using a TWSVM-based classifier and the second problem is an unconstrained minimization problem (UMP) which is reduced to solving a system of linear equations. The second hyperplane is determined so that it is proximal to its own class and the angle between the normals to the two hyperplanes is maximized. The notion of angle has been introduced to have maximum separation between the two hyperplanes. In this work, we have presented two versions of ATWSVM: one that solves a QPP and a UMP; second which formulates both the problems as UMPs. The training time of ATWSVM is much less than that of TWSVM because ATWSVM solves the second problem as UMP instead of QPP. To test the efficacy of the proposed classifier, experiments have been conducted on synthetic and benchmark datasets and it is observed that the proposed classifier achieves classification accuracy comparable or better than that of TWSVM. This work also proposes application of ATWSVM for color image segmentation.

Tree-based localized fuzzy twin support vector clustering with square loss function

Twin support vector machine (TWSVM) is an efficient supervised learning algorithm, proposed for the classification problems. Motivated by its success, we propose Tree-based localized fuzzy twin support vector clustering (Tree-TWSVC). Tree-TWSVC is a novel clustering algorithm that builds the cluster model as a binary tree, where each node comprises of proposed TWSVM-based classifier, termed as localized fuzzy TWSVM (LF-TWSVM). The proposed clustering algorithm Tree-TWSVC has efficient learning time, achieved due to the tree structure and the formulation that leads to solving a series of systems of linear equations. Tree-TWSVC delivers good clustering accuracy because of the square loss function and it uses nearest neighbour graph based initialization method. The proposed algorithm restricts the cluster hyperplane from extending indefinitely by using cluster prototype, which further improves its accuracy. It can efficiently handle large datasets and outperforms other TWSVM-based clustering methods. In this work, we propose two implementations of Tree-TWSVC: Binary Tree-TWSVC and One-against-all Tree-TWSVC. To prove the efficacy of the proposed method, experiments are performed on a number of benchmark UCI datasets. We have also given the application of Tree-TWSVC as an image segmentation tool.

Nonparallel hyperplane classifiers for multi-category classification

Support vector machines (SVMs) are benchmark developments in the field of machine learning. Recently, various nonparallel hyperplanes classification algorithms (NHCAs) have been proposed, which are comparable in terms of classification accuracy when compared with SVM but are computationally more efficient. All these NHCAs are originally proposed for binary classification problems. Since, most of the real world classification problems deal with multiple classes, these algorithms are extended in multi-category scenario. We present a comparative study of four NHCAs - Twin SVM (TWSVM), Generalized eigenvalue proximal SVM (GEPSVM), Regularized GEPSVM (RegGEPSVM) and Improved GEPSVM (IGEPSVM) for multi-category classification. The multi-category classification algorithms for NHCA classifiers are implemented using One-Against-All (OAA), binary tree-based (BT) and ternary decision structure (TDS) approaches and the experiments are performed on benchmark UCI datasets. The experimental results show that TDS-TWSVM outperforms other algorithms concerning classification accuracy and BT-RegGEPSVM takes minimum time for building the classifier.

Enhanced linear block algorithm with improved similarity measure

Content Based Image Retrieval (CBIR) is a technique of finding appropriate images based on the features that are automatically extracted from the image itself. An important low-level feature in any image is dominant color. Dominant Color Descriptor (DCD) was proposed by MPEG-7 and is extensively used in image retrieval. An improvement over DCD was Linear Block Algorithm (LBA). In this paper, we propose an improved similarity measure for dominant color descriptor. We improve LBA by making two significant changes. First is improvement in the similarity measure and second is local implementation through region based dominant colors. The proposed similarity measure takes into account the number of dominant colors of the two images to be compared. The earlier well known methods like MPEG-7 DCD and LBA use the RGB color components and their percentages to find similarity between the query and target images. In our work, it is now weighted by the number of dominant colors in the two images and their mutual distances. The experimental results demonstrate that the proposed method outperforms LBA and other prominent color based retrieval techniques.

Angle-based twin parametric-margin support vector machine for pattern classification

Abstract In this paper, a novel angle-based twin parametric-margin support vector machine (ATP-SVM) is proposed, which can efficiently handle heteroscedastic noise. Taking motivation from twin parametric-margin support vector machine (TPMSVM), ATP-SVM determines two nonparallel parametric-margin hyperplanes, such that the angle between their normal is maximized. Unlike TPMSVM, it solves only one modified quadratic programming problem (QPP) with fewer number of representative samples. Further, it avoids the explicit computation of inverse of matrices in the dual and has efficient learning time as compared to other single problem classifiers like nonparallel SVM based on one optimization problem (NSVMOOP). The efficacy of ATP-SVM is tested by conducting experiments on a wide range of benchmark UCI datasets. ATP-SVM is extended for multi-category classification using state-of-the-art one-against-all (OAA) and binary tree (BT) based multi-category classification approaches. This work also proposes the application of ATP-SVM for segmentation of color images.

Divide and conquer approach for semi-supervised multi-category classification through localized kernel spectral clustering

In this paper, we propose divide-and-conquer approach for multi-category semi-supervised (DAC-MSS) classification and a novel semi-supervised binary classifier termed as twin support vector machine with localized kernel spectral clustering (TW-LKSC). DAC-MSS builds a multi-category classifier model organized in the form of a tree of binary classifiers. The tree consists of several TW-LKSC classifiers which use a training set consisting of few labeled samples and rest unlabeled samples to generate a pair of hyperplanes, by solving a system of linear equations. The propagation of labels to unlabeled patterns is achieved through localized kernel spectral clustering (LKSC) which is the core clustering model embedded in TW-LKSC. TW-LKSC also employs cluster prototype to localize the generation of hyperplanes and prevents them from extending infinitely. The strength of DAC-MSS is its better classification accuracy and improved learning time, due to divide and conquer approach, as compared to one-against-all based semi-supervised classification algorithms. This is proved experimentally for benchmark UCI datasets. We have applied DAC-MSS for color image segmentation of images from Berkley Segmentation Dataset.