Nachimuthu Maheswari
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Nachimuthu Maheswari has a total of 64 co-authors. Here are the top co-authors by number of shared publications:
- Adil Al-Azzawi
Electrical Engineering and Computer Science Department, University of Missouri, Columbia, MO, 65211, USA
2 shared publications Andrews Samraj
Department of Information Technology, Mahendra Engineering College, Namakkal, Tamil Nadu, India
2 shared publicationsK. Duraiswamy
K.S. R. College of Technology
2 shared publicationsM Sivagami
SCSE, Vellore Institute of Technology, Chennai, India
2 shared publicationsM. Sivagami
School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, India
2 shared publications- Sharath Kumar Jagannathan
Saint Peter’s University
2 shared publications
Latest Publications
- Open Access
- 11 November 2026
- Published by Frontiers Media SA in Frontiers in Medical Technology
- Vol. 8, 1957443
HER2 amplification assessment in breast cancer in the HER2-SISH histopathological images using deep learning models
Human Epidermal Growth Factor Receptor 2 (HER2) status is a critical determinant of breast cancer prognosis and therapeutic strategy. Silver in situ Hybridization (SISH) enables HER2 gene assessment under bright-field microscopy while preserving tissue morphology, rendering it a clinically favoured alternative to fluorescence-based methods. Nevertheless, manual SISH slide interpretation is subjective, time-intensive, and susceptible to inter-observer variability, necessitating robust automated computational solutions. This work presents a systematic comparative evaluation of six deep learning architectures for multiclass HER2-SISH histopathology image classification across three diagnostic categories: Amplified, Non-Amplified, and ROI_Normal. The evaluated architectures span convolutional neural networks (MobileNetV3-Large, DenseNet121), transformer-based models (ViT-Small, DeiT-Small), and novel hybrid CNN-Transformer frameworks (DenseNet121-ViT, DenseNet121-DeiT). All models were trained under five-fold stratified cross-validation with a split-before-augmentation protocol and evaluated on an independent held-out test set over both the original and an augmented version of the publicly available HER2-SISH40x dataset. On the original dataset, DenseNet121 achieved the highest test accuracy of 82.10% with a macro F1-score of 77.33%. After offline rotational augmentation of the training partition, DenseNet121 achieved the best performance with test accuracy of 85.09% and macro F1-score of 80.61%, with ViT-Small (84.47%) and DenseNet121-ViT (83.23%) following closely. Per-class analysis revealed that ROI_Normal was consistently classified with the highest accuracy, whereas Amplified versus Non-Amplified discrimination remained the principal challenge across all evaluated models.
- 11 November 2026
- Published by Frontiers Media SA in Frontiers in Medical Technology
- Vol. 8, 1957443
Open AccessHER2 amplification assessment in breast cancer in the HER2-SISH histopathological images using deep learning models
Human Epidermal Growth Factor Receptor 2 (HER2) status is a critical determinant of breast cancer prognosis and therapeutic strategy. Silver in situ Hybridization (SISH) enables HER2 gene assessment under bright-field microscopy while preserving tissue morphology, rendering it a clinically favoured alternative to fluorescence-based methods. Nevertheless, manual SISH slide interpretation is subjective, time-intensive, and susceptible to inter-observer variability, necessitating robust automated computational solutions. This work presents a systematic comparative evaluation of six deep learning architectures for multiclass HER2-SISH histopathology image classification across three diagnostic categories: Amplified, Non-Amplified, and ROI_Normal. The evaluated architectures span convolutional neural networks (MobileNetV3-Large, DenseNet121), transformer-based models (ViT-Small, DeiT-Small), and novel hybrid CNN-Transformer frameworks (DenseNet121-ViT, DenseNet121-DeiT). All models were trained under five-fold stratified cross-validation with a split-before-augmentation protocol and evaluated on an independent held-out test set over both the original and an augmented version of the publicly available HER2-SISH40x dataset. On the original dataset, DenseNet121 achieved the highest test accuracy of 82.10% with a macro F1-score of 77.33%. After offline rotational augmentation of the training partition, DenseNet121 achieved the best performance with test accuracy of 85.09% and macro F1-score of 80.61%, with ViT-Small (84.47%) and DenseNet121-ViT (83.23%) following closely. Per-class analysis revealed that ROI_Normal was consistently classified with the highest accuracy, whereas Amplified versus Non-Amplified discrimination remained the principal challenge across all evaluated models.
- Open Access
- 19 October 2026
- Published by Frontiers Media SA in Frontiers in Big Data
- Vol. 9, 1933830
A retrieval-augmented multi-stage prompting framework for legal analysis in the Indian judicial system
The aim of this endeavor was to create an Artificial Intelligence (AI) based legal researcher that utilizes a multi-stage Retrieval-Augmented Generation (RAG) methodology to solve the challenges of finding and comprehending pertinent legal information in the documents. The traditional methods for finding legal documents will usually yield very few results that are useful to someone without a law degree. This new system uses semantic search, keyword retrieval and Large Language Model (LLM) reasoning methods to improve the legal research process. The submitted query will be transformed into a vector embedding while simultaneously being classified into the appropriate category for retrieval. Once the query is classified, the system will use a hybrid retrieval mechanism to find the most relevant legal document based on both BM25 (Best Matching 25) and FAISS (Facebook AI Similarity Search) search algorithms and will combine these results using Reciprocal Rank Fusion (RRF) in order to determine the best results. Later, it will be processed through LLM models to produce formatted (structured) answers to the query in accordance with the relevant legal citation. The overall performance of the proposed system demonstrates improved answer relevancy, reliability and contextual accuracy, thereby making it an effective tool for legal research and assistance. LLaMA-3.3-70B achieved the highest scores in Answer Relevancy (0.91) and Faithfulness (0.92), proving its ability to generate accurate, intent aligned, and evidence-grounded legal responses with minimal hallucination. Gemini-2.0-Flash showed higher Context Recall (0.78), indicating strong retrieval capability, while Mistral-Large excelled in BLEU (0.769) and semantic similarity, highlighting fluent text generation but weaker evidence grounding.
- 19 October 2026
- Published by Frontiers Media SA in Frontiers in Big Data
- Vol. 9, 1933830
Open AccessA retrieval-augmented multi-stage prompting framework for legal analysis in the Indian judicial system
The aim of this endeavor was to create an Artificial Intelligence (AI) based legal researcher that utilizes a multi-stage Retrieval-Augmented Generation (RAG) methodology to solve the challenges of finding and comprehending pertinent legal information in the documents. The traditional methods for finding legal documents will usually yield very few results that are useful to someone without a law degree. This new system uses semantic search, keyword retrieval and Large Language Model (LLM) reasoning methods to improve the legal research process. The submitted query will be transformed into a vector embedding while simultaneously being classified into the appropriate category for retrieval. Once the query is classified, the system will use a hybrid retrieval mechanism to find the most relevant legal document based on both BM25 (Best Matching 25) and FAISS (Facebook AI Similarity Search) search algorithms and will combine these results using Reciprocal Rank Fusion (RRF) in order to determine the best results. Later, it will be processed through LLM models to produce formatted (structured) answers to the query in accordance with the relevant legal citation. The overall performance of the proposed system demonstrates improved answer relevancy, reliability and contextual accuracy, thereby making it an effective tool for legal research and assistance. LLaMA-3.3-70B achieved the highest scores in Answer Relevancy (0.91) and Faithfulness (0.92), proving its ability to generate accurate, intent aligned, and evidence-grounded legal responses with minimal hallucination. Gemini-2.0-Flash showed higher Context Recall (0.78), indicating strong retrieval capability, while Mistral-Large excelled in BLEU (0.769) and semantic similarity, highlighting fluent text generation but weaker evidence grounding.
- Open Access
- 8 April 2026
- Published by Frontiers Media SA in Frontiers in Built Environment
- Vol. 12, 1790274
Scour depth prediction using machine learning and explainable AI: assessment of bridge vulnerability
Introduction: One of the major factors contributing to the collapse of bridges around the world is bridge scour, whereby sediment is washed away from the bridge foundations by flowing water. Hence, the proper prediction of scour depth is of utmost importance for the proper functioning of the structure and the maintenance schedule. Method: Traditional empirical equations, although they found a wide range of applications, often suffer from their inability to represent the nonlinear interactions between the hydraulic, sediment, and geometric factors. To get through these shortcomings, the present study explores the use of the latest machine learning (ML) models for scour depth prediction based on a combined dataset of 569 experimental cases from the U.S. Geological Survey. A very thorough and considered study was done on Random Forest, Gradient Boosting, XGBoost, CatBoost, Support Vector Regression (SVR), Artificial Neural Networks (ANN), and metaheuristically optimized neural networks (ANN-PSO and ANN-GA). The performance of the models was determined by very comprehensive k-fold cross-validation and evaluated through various statistical indices, such as RRMSE, NMAE, NNSE, KGE, WI, and R 2 . In addition, SHapley Additive exPlanations (SHAP) analysis was used to interpret feature contributions, thereby linking model output to physically consistent hydraulic principles. Results and Discussion: This work discusses the ensemble learning models, significantly Gradient Boosting, provide an accurate and interpretable framework for scour depth prediction. Specifically, Gradient Boosting (five features) achieved an R 2 of 0.9365, RRMSE of 0.2297, NNSE of 0.9407, and WI of 0.9825. By integrating advanced machine learning techniques with explainability tools, the research bridges the gap between black-box modeling and engineering practice. These findings offer a pathway toward the development of reliable decision-support systems for bridge safety assessment, with potential extensions to field-scale validation, temporal scour evolution modeling, and hybrid physics–data-driven approaches. This work analyses the data and interpret the feature contributions using explainable AI, thereby linking model output to physically consistent hydraulic principles. Such prediction of scour depth supports the engineers to take care of the design and safety assessment frameworks for bridges.
- 8 April 2026
- Published by Frontiers Media SA in Frontiers in Built Environment
- Vol. 12, 1790274
Open AccessScour depth prediction using machine learning and explainable AI: assessment of bridge vulnerability
Introduction: One of the major factors contributing to the collapse of bridges around the world is bridge scour, whereby sediment is washed away from the bridge foundations by flowing water. Hence, the proper prediction of scour depth is of utmost importance for the proper functioning of the structure and the maintenance schedule. Method: Traditional empirical equations, although they found a wide range of applications, often suffer from their inability to represent the nonlinear interactions between the hydraulic, sediment, and geometric factors. To get through these shortcomings, the present study explores the use of the latest machine learning (ML) models for scour depth prediction based on a combined dataset of 569 experimental cases from the U.S. Geological Survey. A very thorough and considered study was done on Random Forest, Gradient Boosting, XGBoost, CatBoost, Support Vector Regression (SVR), Artificial Neural Networks (ANN), and metaheuristically optimized neural networks (ANN-PSO and ANN-GA). The performance of the models was determined by very comprehensive k-fold cross-validation and evaluated through various statistical indices, such as RRMSE, NMAE, NNSE, KGE, WI, and R 2 . In addition, SHapley Additive exPlanations (SHAP) analysis was used to interpret feature contributions, thereby linking model output to physically consistent hydraulic principles. Results and Discussion: This work discusses the ensemble learning models, significantly Gradient Boosting, provide an accurate and interpretable framework for scour depth prediction. Specifically, Gradient Boosting (five features) achieved an R 2 of 0.9365, RRMSE of 0.2297, NNSE of 0.9407, and WI of 0.9825. By integrating advanced machine learning techniques with explainability tools, the research bridges the gap between black-box modeling and engineering practice. These findings offer a pathway toward the development of reliable decision-support systems for bridge safety assessment, with potential extensions to field-scale validation, temporal scour evolution modeling, and hybrid physics–data-driven approaches. This work analyses the data and interpret the feature contributions using explainable AI, thereby linking model output to physically consistent hydraulic principles. Such prediction of scour depth supports the engineers to take care of the design and safety assessment frameworks for bridges.
- Open Access
- 2 March 2026
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 17, 1717252
EEFSA-SECM: an enhanced ensemble feature selection and stacking ensemble classifier to detect Parkinson’s disease
Introduction: Parkinson’s disease (PD) is a progressive neurological disorder whose early symptoms often remain undetected, making timely diagnosis challenging. Machine learning offers strong algorithms to detect subtle speech-based biomarkers that are impossible to detect by standard methods. Methods: In this article, we proposed an Enhanced Ensemble Feature Selection Algorithm (EEFSA) which combines filter, wrapper, and embedded approaches to extract the best informative features, eliminate redundancy, and improve classification performance. The proposed work has tested on two benchmark audio based datasets, such as Dataset-1 (46 features, 80 samples), where EEFSA reduced the features to 20 features, and Dataset-2 (754 features, 252 samples), where EEFSA reduced the features to 40 features. Nine machine learning classifiers were tried out and the best of them were combined into a stacking ensemble with logistic regression as the meta-classifier. Results: Experiments show that EEFSA-driven dimensionality reduction not only enhanced accuracy of classification but also reduced training time considerably and minimized over fitting. The Stacking Ensemble Classifier Model (SECM) deployed on the basis of the proposed method achieved accuracy of 86.67% and 89.95% on Dataset-1 and Dataset-2, respectively, and outperformed individual classifiers in all experiments. Conclusion: Overall, this work provides EEFSA-driven stacking as a new and efficient method of feature selection and ensemble learning combination for Parkinson’s disease classification. The proposed EEFSA–SECM framework achieves effective classification accuracy, competitive training/testing times, and improved AUC scores on two benchmark datasets, establishing it as an effective and efficient approach for Parkinson’s disease diagnosis.
- 2 March 2026
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 17, 1717252
Open AccessEEFSA-SECM: an enhanced ensemble feature selection and stacking ensemble classifier to detect Parkinson’s disease
Introduction: Parkinson’s disease (PD) is a progressive neurological disorder whose early symptoms often remain undetected, making timely diagnosis challenging. Machine learning offers strong algorithms to detect subtle speech-based biomarkers that are impossible to detect by standard methods. Methods: In this article, we proposed an Enhanced Ensemble Feature Selection Algorithm (EEFSA) which combines filter, wrapper, and embedded approaches to extract the best informative features, eliminate redundancy, and improve classification performance. The proposed work has tested on two benchmark audio based datasets, such as Dataset-1 (46 features, 80 samples), where EEFSA reduced the features to 20 features, and Dataset-2 (754 features, 252 samples), where EEFSA reduced the features to 40 features. Nine machine learning classifiers were tried out and the best of them were combined into a stacking ensemble with logistic regression as the meta-classifier. Results: Experiments show that EEFSA-driven dimensionality reduction not only enhanced accuracy of classification but also reduced training time considerably and minimized over fitting. The Stacking Ensemble Classifier Model (SECM) deployed on the basis of the proposed method achieved accuracy of 86.67% and 89.95% on Dataset-1 and Dataset-2, respectively, and outperformed individual classifiers in all experiments. Conclusion: Overall, this work provides EEFSA-driven stacking as a new and efficient method of feature selection and ensemble learning combination for Parkinson’s disease classification. The proposed EEFSA–SECM framework achieves effective classification accuracy, competitive training/testing times, and improved AUC scores on two benchmark datasets, establishing it as an effective and efficient approach for Parkinson’s disease diagnosis.
- Open Access
- 19 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 136846-136879
A Comprehensive Review of Recent Advances in Deep Neural Networks for Lipreading With Sign Language Recognition
Lip reading is a form of "listening" to people that happens visually. It’s also referred to as "Speech reading." This is done by observing the speaker’s face and listening to the spoken words. The mechanism comprises face identification, lip localization, feature extraction, and finally identifying the phrase/sentence based on the lip movements. It recognizes the text through lip movements. The expert system has learned by transferring the end user-generated lip-movement frames with the sequence. It will distinguish the lip movement and the uttered word using visual input or audio and video data. This article discusses the improvement of lip reading and the involvement of sign language in lip reading. The feature extraction methods, lip-reading approaches, datasets, parameters, and different languages used are discussed in this article. At the same time, the article also focused on various datasets and approaches for integration of sign language with lip reading. Finally, the challenges of recent lipreading techniques and future research directions are highlighted.
- 19 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 136846-136879
Open AccessA Comprehensive Review of Recent Advances in Deep Neural Networks for Lipreading With Sign Language Recognition
Lip reading is a form of "listening" to people that happens visually. It’s also referred to as "Speech reading." This is done by observing the speaker’s face and listening to the spoken words. The mechanism comprises face identification, lip localization, feature extraction, and finally identifying the phrase/sentence based on the lip movements. It recognizes the text through lip movements. The expert system has learned by transferring the end user-generated lip-movement frames with the sequence. It will distinguish the lip movement and the uttered word using visual input or audio and video data. This article discusses the improvement of lip reading and the involvement of sign language in lip reading. The feature extraction methods, lip-reading approaches, datasets, parameters, and different languages used are discussed in this article. At the same time, the article also focused on various datasets and approaches for integration of sign language with lip reading. Finally, the challenges of recent lipreading techniques and future research directions are highlighted.
- Open Access
- 3 August 2023
- Published by Elsevier in Alexandria Engineering Journal
- Vol. 78, 469-482
An effective technique for detecting minority attacks in NIDS using deep learning and sampling approach
- 3 August 2023
- Published by Elsevier in Alexandria Engineering Journal
- Vol. 78, 469-482
Open AccessAn effective technique for detecting minority attacks in NIDS using deep learning and sampling approach
- 19 March 2023
- Published by Springer Nature
- p. 300-317
Nowcasting - Rain Intensity Prediction Using Machine Learning Models with Markov Chain Process
- 19 March 2023
- Published by Springer Nature
- p. 300-317
Nowcasting - Rain Intensity Prediction Using Machine Learning Models with Markov Chain Process
- 14 May 2020
- Published by Taylor & Francis in International Journal of Computers and Applications
- Vol. 44 (4) , 396-400
A model for screening eye diseases using optical coherence tomography images
Age-Related Macular Degeneration (AMD) is one of the most common type of eye ailments which generally affects elderly people of age above 60 years and which damages their central vision. Diabetic Macular Edema (DME) is an another form of irreversible vision loss that occurs in diabetic people because of the accumulation of fat in retinal nerves. Therefore it is necessary to screen AMD and DME at an early stage, as it would help in the treatment and prevention of complete blindness. This paper aims to create a system for screening AMD and DME on Optical Coherence Tomography images using Image processing techniques in order to improve the feature to be extracted from the images using Convolutional Neural Network.
- 14 May 2020
- Published by Taylor & Francis in International Journal of Computers and Applications
- Vol. 44 (4) , 396-400
A model for screening eye diseases using optical coherence tomography images
Age-Related Macular Degeneration (AMD) is one of the most common type of eye ailments which generally affects elderly people of age above 60 years and which damages their central vision. Diabetic Macular Edema (DME) is an another form of irreversible vision loss that occurs in diabetic people because of the accumulation of fat in retinal nerves. Therefore it is necessary to screen AMD and DME at an early stage, as it would help in the treatment and prevention of complete blindness. This paper aims to create a system for screening AMD and DME on Optical Coherence Tomography images using Image processing techniques in order to improve the feature to be extracted from the images using Convolutional Neural Network.
- Open Access
- 1 March 2020
- Published by Scalable Computing: Practice and Experience in Scalable Computing: Practice and Experience
- Vol. 21 (1) , 17-31
BACKGROUND MODELLING USING A Q-TREE BASED FOREGROUND SEGMENTATION
Background modelling is an empirical part in the procedure of foreground mining of idle and moving objects. The foreground object detection has become a challenging phenomenon due to intermittent objects, intensity variation, image artefact and dynamic background in the video analysis and video surveillance applications. In the video surveillances application, a large amount of data is getting processed by everyday basis. Thus it needs an efficient background modelling technique which could process those larger sets of data which promotes effective foreground detection. In this paper, we presented a renewed background modelling method for foreground segmentation. The main objective of the work is to perform the foreground extraction only in the intended region of interest using proposed Q-Tree algorithm. At most all the present techniques consider their updates to the pixels of the entire frame which may result in inefficient foreground detection with a quick update to slow moving objects; The proposed method contract these defect by extracting the foreground object by controlling the region of interest (the region only where the background subtraction is to be performed) and thereby reducing the false positive and false negative. The extensive experimental results and the evaluation parameters of the proposed approach with the state of art method were compared against the most recent background subtraction approaches. Moreover, we use challenge change detection dataset and the efficiency of our method is analyzed in different environmental conditions (indoor, outdoor) from the CDnet2014 dataset and additional real time videos. The experimental results were satisfactorily verified the strengths and weakness of proposed method against the existing state-of-the-art background modelling methods.
- 1 March 2020
- Published by Scalable Computing: Practice and Experience in Scalable Computing: Practice and Experience
- Vol. 21 (1) , 17-31
Open AccessBACKGROUND MODELLING USING A Q-TREE BASED FOREGROUND SEGMENTATION
Background modelling is an empirical part in the procedure of foreground mining of idle and moving objects. The foreground object detection has become a challenging phenomenon due to intermittent objects, intensity variation, image artefact and dynamic background in the video analysis and video surveillance applications. In the video surveillances application, a large amount of data is getting processed by everyday basis. Thus it needs an efficient background modelling technique which could process those larger sets of data which promotes effective foreground detection. In this paper, we presented a renewed background modelling method for foreground segmentation. The main objective of the work is to perform the foreground extraction only in the intended region of interest using proposed Q-Tree algorithm. At most all the present techniques consider their updates to the pixels of the entire frame which may result in inefficient foreground detection with a quick update to slow moving objects; The proposed method contract these defect by extracting the foreground object by controlling the region of interest (the region only where the background subtraction is to be performed) and thereby reducing the false positive and false negative. The extensive experimental results and the evaluation parameters of the proposed approach with the state of art method were compared against the most recent background subtraction approaches. Moreover, we use challenge change detection dataset and the efficiency of our method is analyzed in different environmental conditions (indoor, outdoor) from the CDnet2014 dataset and additional real time videos. The experimental results were satisfactorily verified the strengths and weakness of proposed method against the existing state-of-the-art background modelling methods.
- 31 January 2020
- Published by Emerald Publishing in Information Discovery and Delivery
- Vol. 48 (1) , 23-34
An effective foreground segmentation using adaptive region based background modelling
Purpose: Background modelling has played an imperative role in the moving object detection as the progress of foreground extraction during video analysis and surveillance in many real-time applications. It is usually done by background subtraction. This method is uprightly based on a mathematical model with a fixed feature as a static background, where the background image is fixed with the foreground object running over it. Usually, this image is taken as the background model and is compared against every new frame of the input video sequence. In this paper, the authors presented a renewed background modelling method for foreground segmentation. The principal objective of the work is to perform the foreground object detection only in the premeditated region of interest (ROI). The ROI is calculated using the proposed algorithm reducing and raising by half (RRH). In this algorithm, the coordinate of a circle with the frame width as the diameter is considered for traversal to find the pixel difference. The change in the pixel intensity is considered to be the foreground object and the position of it is determined based on the pixel location. Most of the techniques study their updates to the pixels of the complete frame which may result in increased false rate; The proposed system deals these flaw by controlling the ROI object (the region only where the background subtraction is performed) and thus extracts a correct foreground by exactly categorizes the pixel as the foreground and mines the precise foreground object. The broad experimental results and the evaluation parameters of the proposed approach with the state of art methods were compared against the most recent background subtraction approaches. Moreover, the efficiency of the authors’ method is analyzed in different situations to prove that this method is available for real-time videos as well as videos available in the 2014 challenge change detection data set. Design/methodology/approach: In this paper, the authors presented a fresh background modelling method for foreground segmentation. The main objective of the work is to perform the foreground object detection only on the premeditated ROI. The region for foreground extraction is calculated using proposed RRH algorithm. Most of the techniques study their updates to the pixels of the complete frame which may result in increased false rate; most challenging case is that, the slow moving object is updated quickly to detect the foreground region. The anticipated system deals these flaw by controlling the ROI object (the region only where the background subtraction is performed) and thus extracts a correct foreground by exactly categorizing the pixel as the foreground and mining the precise foreground object. Findings: Plum Analytics provide a new conduit for documenting and contextualizing the public impact and reach of research within digitally networked environments. While limitations are notable, the metrics promoted through the platform can be used to build a more comprehensive view of research impact. Originality/value: The algorithm used in the work was proposed by the authors and are used for experimental evaluations.
- 31 January 2020
- Published by Emerald Publishing in Information Discovery and Delivery
- Vol. 48 (1) , 23-34
An effective foreground segmentation using adaptive region based background modelling
Purpose: Background modelling has played an imperative role in the moving object detection as the progress of foreground extraction during video analysis and surveillance in many real-time applications. It is usually done by background subtraction. This method is uprightly based on a mathematical model with a fixed feature as a static background, where the background image is fixed with the foreground object running over it. Usually, this image is taken as the background model and is compared against every new frame of the input video sequence. In this paper, the authors presented a renewed background modelling method for foreground segmentation. The principal objective of the work is to perform the foreground object detection only in the premeditated region of interest (ROI). The ROI is calculated using the proposed algorithm reducing and raising by half (RRH). In this algorithm, the coordinate of a circle with the frame width as the diameter is considered for traversal to find the pixel difference. The change in the pixel intensity is considered to be the foreground object and the position of it is determined based on the pixel location. Most of the techniques study their updates to the pixels of the complete frame which may result in increased false rate; The proposed system deals these flaw by controlling the ROI object (the region only where the background subtraction is performed) and thus extracts a correct foreground by exactly categorizes the pixel as the foreground and mines the precise foreground object. The broad experimental results and the evaluation parameters of the proposed approach with the state of art methods were compared against the most recent background subtraction approaches. Moreover, the efficiency of the authors’ method is analyzed in different situations to prove that this method is available for real-time videos as well as videos available in the 2014 challenge change detection data set. Design/methodology/approach: In this paper, the authors presented a fresh background modelling method for foreground segmentation. The main objective of the work is to perform the foreground object detection only on the premeditated ROI. The region for foreground extraction is calculated using proposed RRH algorithm. Most of the techniques study their updates to the pixels of the complete frame which may result in increased false rate; most challenging case is that, the slow moving object is updated quickly to detect the foreground region. The anticipated system deals these flaw by controlling the ROI object (the region only where the background subtraction is performed) and thus extracts a correct foreground by exactly categorizing the pixel as the foreground and mining the precise foreground object. Findings: Plum Analytics provide a new conduit for documenting and contextualizing the public impact and reach of research within digitally networked environments. While limitations are notable, the metrics promoted through the platform can be used to build a more comprehensive view of research impact. Originality/value: The algorithm used in the work was proposed by the authors and are used for experimental evaluations.