Anandarup Roy

JCLMM: A finite mixture model for clustering of circular-linear data and its application to psoriatic plaque segmentation

Abstract The hue and chroma components of an image pixel carry crucial information that can be exploited to perform segmentation. However, due to its directional property, a circular distribution is required to characterize the hue component. In this article, we propose a mixture of bi-variate circular–linear distributions, for modelling hue and chroma information. The proposed model incorporates a joint distribution of a circular and a linear variable by means of circular copula and offers a flexible architecture that deals with heterogeneous margins for different mixture components. We apply this model for psoriatic plaque segmentation in skin images, using the hue and the chroma information. We observe that the chroma exhibits a heterogeneous distribution in a skin image. Moreover, the joint distribution of hue and chroma possesses multi-modal characteristics. Our model is suitable to perform segmentation under such circumstances. After segmentation, we perform automatic plaque localization by means of a statistical model that exploits hue information of the segmented regions. We conduct the experiments on a set of 75 psoriasis skin images. Both segmentation and localization performances are evaluated with respect to a number of commonly used criteria. The experimental results show that the proposed segmentation model outperforms several competing supervised and unsupervised methods in detecting psoriatic plaque regions in skin images.

A study on combining dynamic selection and data preprocessing for imbalance learning

Abstract In real life, classifier learning may encounter a dataset in which the number of instances of a given class is much higher than for other classes. Such imbalanced datasets require special attention because traditional classifiers generally favor the majority class which has a large number of instances. Ensemble classifiers, in such cases, have been reported to yield promising results. Most often, ensembles are specially designed for data level preprocessing techniques that aim to balance class proportions by applying under-sampling and/or over-sampling. Most available studies concentrate on static ensembles designed for different preprocessing techniques. Contrary to static ensembles, dynamic ensembles became popular thanks to their performance in the context of ill defined problems (small size datasets). A dynamic ensemble includes a dynamic selection module for choosing the best ensemble given a test instance. This paper experimentally evaluates the argument that dynamic selection combined with a preprocessing technique can achieve higher performance than static ensemble for imbalanced classification problems. For this evaluation, we collect 84 two-class and 26 multi-class datasets of varying degrees of class-imbalance. In addition, we consider five variations of preprocessing methods and four dynamic selection methods. We further design a useful experimental framework to integrate preprocessing and dynamic selection. Our experiments show that the dynamic ensemble improves the F-measure and the G-mean as compared to the static ensemble. Moreover, considering different levels of imbalance, dynamic selection methods secure higher ranks than other alternatives.

Meta-learning recommendation of default size of classifier pool for META-DES

Abstract Dynamic ensemble selection (DES) is a mechanism for selecting an ensemble of competent classifiers from a pool of base classifiers, in order to classify a particular test sample. The size of this pool is user-defined, and yet is crucial for controlling the computational complexity and performance of a DES. An appropriate pool size depends on the choice of base classifiers, the underlying DES method used, and more importantly, the characteristics of the given problem. After the DES method and the base classifiers are selected, an appropriate pool size for a given problem can be obtained by the repetitive application of the DES with a variety of sizes, after which a selection is performed. Since this brute force approach is computationally expensive, researchers set the pool size to a pre-specified value. This strategy, may, however further complicate and reduce the performance of the DES method. Instead, we propose a framework that is akin to meta-learning, in order to predict a suitable pool size based on the intrinsic classification complexity of a problem. In our strategy, we collect meta-features corresponding to classification complexity from a number of data sets. Additionally, we obtain the best pool sizes for these data sets using the brute force approach. The association between these two pieces of information is captured using meta-regression models. Finally, for an unseen problem, we predict the pool size using this model and the classification complexity information. We carry out experiments on 65 two-class data sets and with a recent DES method, namely, META-DES. We also consider variants of meta-regression techniques and report prediction results, after which we carry out a statistical comparison among them. Moreover, we investigate the performance of META-DES and observe that it performs equivalently for both the predicted and the best pool sizes.

Script independent online handwriting recognition

The most general form of handwriting style is mixed cursive and this is the most difficult type in view of its automatic recognition. Similar handwriting style are prevalent in various scripts such as English, Arabic, Bengali etc. Handwriting recognition for such a script gets further difficult whenever its alphabet consists of a large number of characters like Bengali which has around 350 characters. Hidden Markov models (HMM) are the most popularly used architectures for similar recognition problems. However, the task becomes easy if the underlying lexicon depending upon the specific application is provided. In such situations, holistic or word-based recognition approach is adopted which does not require recognition of the constituent characters. On the other hand, the same task gets complicated as the lexicon size increases and / or it consists of many similar shape words. In a recent study [1] of similar situation, a fully connected non-homogeneous HMM has been used where its observation sequence was generated through explicit segmentation of the input word. In the present study, we have explored that the performance of this HMM-based recognition scheme is independent of both the script and the particular intelligent segmentation strategy. We implemented a novel segmentation scheme based on Discrete Curve Evolution algorithm [2] and two other existing segmentation methods on standard databases of English, Arabic and Bangla to arrive at the above conclusion. Statistical hypothesis testings of the simulation results further confirm the above claim.

Mixture model based color clustering for psoriatic plaque segmentation

This paper presents a mixture model based color clustering and then applies this technique for psoriatic plaque segmentation in skin images. For clustering image pixels, two mostly relevant colorspaces namely, CIE Luv(cubic) and CIE Lch(equivalent cylindrical) are considered. Gaussian Mixture Model(GMM) is used for clustering in Luv space. However, Lch space being a circular-linear space does not support the use of GMM. Hence, clustering in Lch makes use of a novel mixture model known as Semi-Wrapped Gaussian Mixture Model(SWGMM). The performance of these clustering methods is evaluated for psoriatic plaque segmentation and results are compared with those obtained by the commonly used Fuzzy C-Means (FCM) clustering algorithm. The comparative study shows that the clustering in Lch using SWGMM outperforms the other approaches. For localizing the plaques, we consider von Mises distribution to find a suitable confidence interval and thereby defining skin and non-skin models. The UCI Skin Segmentation dataset is used for this purpose. This localization approach achieves an average accuracy 79.53%. A real clinical dataset of Psoriasis images is used in this experiment.

Meta-regression based pool size prediction scheme for dynamic selection of classifiers

Dynamic selection (DS) is a mechanism to select one or an ensemble of competent classifiers from a pool of base classifiers, in order to classify a specific test sample. The size of this pool is user defined and yet crucial to control the computational complexity and performance of a DS. An appropriate pool size depends on the choice of base classifiers, the underlying DS method used, and more importantly, the characteristics of the given problem. After the DS method and the base classifiers are selected, an appropriate pool size for a given problem can be obtained by the repetitive application of the DS with a variety of sizes, after which a selection is performed. Since this brute force approach is computationally expensive, researchers usually set the pool size to a pre-specified value. However, this strategy may reduce the performance of the DS method. Instead, we propose a meta-regression model in order to predict a suitable pool size, based on the intrinsic classification complexity of a problem. In our strategy, we obtain the best pool sizes for a number of data sets, using the brute force approach. Additionally, we extract meta-features that represent classification complexity of a problem. These two pieces of information are associated by means of meta-regression models. Finally, for an unseen problem, we predict the pool size using this model and the classification complexity information.We carry out the experiments on 64 two-class data sets and with several well-known DS methods. We also consider variants of meta-regression techniques and report prediction results. We further analyze these results using a statistical test. Finally, we investigate the performance of a DS and observe that DS performs equivalently for predicted and the best pool sizes.

A Mixture Model of Circular-Linear Distributions for Color Image Segmentation

This article deals with mixture model based color image segmentation in the LCH color space. In this space, one of the components (representing hue in particular) is circular in nature. Hence LCH image pixels are samples on a cylinder. A statistical model for such data needs to employ circular-linear joint distributions. Here such a model is designed using the “Independent von-Mises Gaussian” distribution. Further its mixture is used to approximate the distribution of the LCH data. The mixture parameters are estimated using standard EM algorithm. Comprehensive experiments are conducted on Berkeley segmentation data set to measure the performance of the algorithm in terms of a variety of quantitative indices for image segmentation. A comparison is further made with some existing mixture models. Our study reveals that the proposed mixture model performs satisfactorily in this regard.

Color Based Image Segmentation

This article addresses color image segmentation in hue-saturation space. A model for circular data is provided by the vM-Gauss distribution, which is a joint distribution of von-Mises and Gaussian distribution. The mixture of vM-Gauss distribution is used to model hue-saturation data. A cluster merging process is applied to separate such identifiable objects in the image. The results are shown on Berkeley segmentation dataset. A cluster association methodology is developed for comparison.

An HMM framework based on spherical-linear features for online cursive handwriting recognition

Abstract In this paper a Hidden Markov Model (HMM) based writer independent online unconstrained handwritten word recognition scheme is proposed. The main steps here are segmentation of handwritten word samples into sub-strokes, feature extraction from the sub-strokes and recognition. We propose a novel but simple strategy based on the well-known discrete curve evolution for the segmentation task. Next, certain angular and linear features are extracted from the sub-strokes of word samples and are modelled as feature vectors generated from a mixture distribution. This mixture model is designed to accommodate the correlation among the angular variables. We formulate a Baum-Welch parameter estimation algorithm that can handle spherical-linear correlated data to construct an HMM. Finally, based on this HMM, we design a classifier for recognition of handwritten word samples. Simulation trials have been conducted on handwritten word sample databases of Latin and Bangla scripts demonstrating successful performance of the proposed recognition scheme.