Showing papers by "Ishwar K. Sethi published in 2015"

Blind components processing a novel approach to array signal processing: A research orientation

Abstract: Blind Components Processing (BCP), a novel approach in processing of data (signal, image, etc.) components, is introduced as well some applications to information communications technology (ICT) are proposed. The newly introduced BCP is with capability of deployment orientation in a wider range of applications. The fundamental of BCP is based on Blind Source Separation (BSS), a methodology which searches for unknown sources of mixtures without a prior knowledge of either the sources or the mixing process. Most of the natural, biomedical as well as industrial observed signals are mixtures of different components while the components and the way they mixed are unknown. If we decompose the signal into its components by BSS, then we can process the components separately without interfering the other components signal/data. Such internal access to signal components leads to extraction of plenty of information as well more efficient processing compared to normal signal processing wherein all the structure of the signal is gone under processing and modification. This manuscript besides the introducing BCP, proposes a practical applications of BCP with technical merit for harmonic noise cancellation as well stock pricing model.

Soft-Hard Clustering for Multiview Data

Abstract: With rapid advances in technology and connectivity, the capability to capture data from multiple sources has given rise to multiview learning wherein each object has multiple representations and a learned model, whether supervised or unsupervised, needs to integrate these different representations. Multiview learning has shown to yield better predictive and clustering models, it also is able to provide a better insight into relationships between different views for making better decisions. In this paper, we consider the problem of multiview clustering and present a soft-hard clustering approach. In our approach, all object views are first independently mapped into a unit hypercube via soft clustering. The mapped views are next integrated via a hard clustering approach to yield the final results. Both soft and hard clustering stages utilize k-means or its variant c-means, which makes our method suitable for large-scale data problems. Furthermore, additional parallelization of the view mapping stage in parallel is possible, thus making the method more attractive for large-scale data applications. The performance of the method using three benchmark data sets is demonstrated and a comparison with other published results shows our method mostly yields a slightly better performance.

A Zoned Image Patch Permutation Descriptor

Abstract: Image representation through local descriptors is a research hotspot in computer vision. In this letter, we propose a novel local image descriptor based on intensity permutation and zone division. The oFAST detector is first employed to detect keypoints with orientations, and then steered patterns are applied to sample rotation-invariant points within the local keypoint patch. In the step of local patch description, intensity permutation and zone division are implemented to construct our descriptor, with the advantages of inherent robustness and invariance to monotonic brightness changes. Our proposed algorithm performed well in the experiments on benchmark dataset for descriptor evaluation.

Multi-label Classification with a Constrained Minimum Cut Model

Abstract: Multi-label classification has received more attention recently in the fields of data mining and machine learning. Though many approaches have been proposed, the critical issue of how to combine single labels to form a multi-label remains challenging. In this work, we propose a novel multi-label classification approach that each label is represented by two exclusive events: the label is selected or not selected. Then a weighted graph is used to represent all the events and their correlations. The multi-label learning is transformed into finding a constrained minimum cut of the weighted graph. In the experiments, we compare the proposed approach with the state-of-the-art multi-label classifier ML-KNN, and the results show that the new approach is efficient in terms of all the popular metrics used to evaluate multi-label classification performance.