Interest point detection using imbalance oriented selection

NeuroImage, 2011

Learning with discriminative methods is generally based on minimizing the misclassification of training samples, which may be unsuitable for imbalanced datasets where the recognition might be biased in favor of the most numerous class. This problem can be addressed with a generative approach, which typically requires more parameters to be determined leading to reduced performances in high dimension. In such situations, dimension reduction becomes a crucial issue. We propose a feature selection/classification algorithm based on generative methods in order to predict the clinical status of a highly imbalanced dataset made of PET scans of forty-five low-functioning children with autism spectrum disorders (ASD) and thirteen non-ASD low functioning children. ASDs are typically characterized by impaired social interaction, narrow interests, and repetitive behaviors, with a high variability in expression and severity. The numerous findings revealed by brain imaging studies suggest that ASD is associated with a complex and distributed pattern of abnormalities that makes the identification of a shared and common neuroimaging profile a difficult task. In this context, our goal is to identify the rest functional brain imaging abnormalities pattern associated with ASD and to validate its efficiency in individual classification. The proposed feature selection algorithm detected a characteristic pattern in the ASD group that included a hypoperfusion in the right Superior Temporal Sulcus (STS) and a hyperperfusion in the contralateral postcentral area. Our algorithm allowed for a significantly accurate (88%), sensitive (91%) and specific (77%) prediction of clinical category. For this imbalanced dataset, with only 13 control scans, the proposed generative algorithm outperformed other state-of-the-art discriminant methods. The high predictive power of the characteristic pattern, which has been automatically identified on whole brains without any priors, confirms previous findings concerning the role of STS in ASD. This work offers exciting possibilities for early autism detection and/or the evaluation of treatment response in individual patients.

International Journal of …, 2011

Since dealing with high dimensional data is computationally complex and sometimes even intractable, recently several feature reduction methods have been developed to reduce the dimensionality of the data in order to simplify the calculation analysis in various applications such as text categorization, signal processing, image retrieval and gene expressions among many others. Among feature reduction techniques, feature selection is one of the most popular methods due to the preservation of the original meaning of features. However, most of the current feature selection methods do not have a good performance when fed on imbalanced data sets which are pervasive in real world applications. In this paper, we propose a new unsupervised feature selection method attributed to imbalanced data sets, which will remove redundant features from the original feature space based on the distribution of features. To show the effectiveness of the proposed method, popular feature selection methods have been implemented and compared. Experimental results on the several imbalanced data sets, derived from UCI repository database, illustrate the effectiveness of the proposed method in comparison with other rival methods in terms of both AUC and F1 performance measures of 1-Nearest Neighbor and Naïve Bayes classifiers and the percent of the selected features.

IEEE/CAA Journal of Automatica Sinica, 2019

Imbalanced data is one type of datasets that are frequently found in real-world applications, e.g., fraud detection and cancer diagnosis. For this type of datasets, improving the accuracy to identify their minority class is a critically important issue. Feature selection is one method to address this issue. An effective feature selection method can choose a subset of features that favor in the accurate determination of the minority class. A decision tree is a classifier that can be built up by using different splitting criteria. Its advantage is the ease of detecting which feature is used as a splitting node. Thus, it is possible to use a decision tree splitting criterion as a feature selection method. In this paper, an embedded feature selection method using our proposed weighted Gini index (WGI) is proposed. Its comparison results with Chi2, F-statistic and Gini index feature selection methods show that F-statistic and Chi2 reach the best performance when only a few features are selected. As the number of selected features increases, our proposed method has the highest probability of achieving the best performance. The area under a receiver operating characteristic curve (ROC AUC) and F-measure are used as evaluation criteria. Experimental results with two datasets show that ROC AUC performance can be high, even if only a few features are selected and used, and only changes slightly as more and more features are selected. However, the performance of F-measure achieves excellent performance only if 20% or more of features are chosen. The results are helpful for practitioners to select a proper feature selection method when facing a practical problem.

Applied Intelligence, 2019

Class imbalance is one of the critical areas in classification. The challenges become more severe when the data set has a large number of features. Traditional classifiers generally favour the majority class because of skewed class distributions. In recent years, feature selection is being used to select the appropriate features for better classification of minority class. However, these studies are limited to imbalance that arise between the classes. In addition to between class imbalance, within class imbalance, along with large number of features, adds additional complexity and results in poor performance of the classifier. In the current study, we propose an effective distance based feature selection method (ED-Relief) that uses a sophisticated distance measure, in order to tackle simultaneous occurrence of between and within class imbalance. This method has been tested on a variety of simulated experiments and real life data sets and the results are compared with the traditional Relief method and some of the well known recent distance based feature selection methods. The results clearly show the superiority of the proposed effective distance based feature selection method.

2011 International Conference on Computer Vision, 2011

In this paper, we describe an interest point detector using edge foci. Unlike traditional detectors that compute interest points directly from image intensities, we use normalized intensity edges and their orientations. We hypothesize that detectors based on the presence of oriented edges are more robust to non-linear lighting variations and background clutter than intensity based techniques. Specifically, we detect edge foci, which are points in the image that are roughly equidistant from edges with orientations perpendicular to the point. The scale of the interest point is defined by the distance between the edge foci and the edges. We quantify the performance of our detector using the interest point's repeatability, uniformity of spatial distribution, and the uniqueness of the resulting descriptors. Results are found using traditional datasets and new datasets with challenging non-linear lighting variations and occlusions.

2021

Applications for interest point detectors and descriptors are just as broad as the field of computer vision as a whole. Each technique has a slightly different approach to cover its niche. With the growing number of new detectors and descriptors, it is crucial to evaluate in which context the detector is most potent and where its struggles lie. In this paper, we will evaluate a collection of novel and established local detectors and descriptors. First, we design a dataset so that the methods are exposed to a diverse set of conditions. With this dataset, the stability of interest point detectors is tested, after which the matching capabilities of the descriptors are put to the test. The methods will also be evaluated on their capability to detect copies from a large set of images. Finally, an existing method will be improved to make it more robust to transformations. This study aims to help researchers identify which method is best appropriate for their purposes.

DOI to the publisher's website. • The final author version and the galley proof are versions of the publication after peer review. • The final published version features the final layout of the paper including the volume, issue and page numbers. Link to publication General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal. If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the "Taverne" license above, please follow below link for the End User Agreement:

2006

Lecture Notes in Computer Science, 2010

Region Of Interest Based Image Classification (ROIBIC) is a mechanism for categorising images according to some specific component or object that features across a given image set. This paper describes and compares two such approaches. The first is founded on a weighted graph mining technique whereby the ROI is represented using a tree structure which allows the application of a weighted graph mining technique to identify features of interest, which can then be used as the foundation with which to build a classifier. The second approach is founded on a time series analysis technique whereby the ROI are represented as time series which can then be used as the foundation for a Case Based Reasoner. The presented evaluation focuses on MRI brain scan data where the classification is focused on the corpus callosum, a distinctive region in MRI brain scan data. Two scenarios are considered: distinguishing between musicians and non-musicians and epilepsy patient screening.

2013

Each data set with imbalanced distribution in its classes can be considered as imbalanced data. Text classification, image classification, clustering web pages and risk management are only part common uses of these data sets. Different methods for processing imbalanced data have been proposed. Among these methods, feature selection is one of the newest and most effective methods. The goal of feature selection methods is selecting the subset of features that the classification process has the best performance. Feature selection methods include wrapper, embedded, and the filtering methods. Between feature selection methods, filtering method is one of the best methods for imbalanced data processing. For this reason, in this paper a new method for imbalanced data processing that is suitable for imbalanced datasets with small number of samples and high number of features is proposed. In this approach, In order to evaluate the value of each feature, the distribution function of each class...