Computer Vision, Medical Image Analysis, Machine Learning, Deep Learning & Parallel Computing.
Current Research Activities
Ph.D. Thesis Work | December 2021 – Present
Title: Deep Learning for Automated Skin Lesion Analysis: Methods and Applications
Thesis Supervisor: Dr. Surya Prakash, Professor, Department of Computer Science and Engineering, IIT Indore, India.
Research Areas: Medical Image Analysis, Skin Lesion Classification, Deep Learning, Representation Learning, Attention Mechanisms, Explainable AI, and Responsible AI.
Research Objective: To develop robust, efficient, and generalizable deep-learning methods for automated skin lesion classification, with an emphasis on improving diagnostic reliability, addressing class imbalance and domain variability, and enabling computationally efficient deployment.
Research Description: My Ph.D. research focuses on developing advanced deep-learning methodologies for automated skin lesion analysis and classification. The primary objective is to improve the reliability, robustness, and generalizability of computer-aided skin lesion assessment across diverse datasets and imaging conditions. The research investigates key challenges in dermatological image analysis, including class imbalance, inter-class visual similarity, intra-class variability, differences in lesion morphology, color and texture, imaging artifacts, and cross-dataset domain shifts.
The proposed research explores complementary approaches based on adaptive feature learning, attention mechanisms, contrastive representation learning, structured feature selection, and data augmentation optimization. Particular emphasis is placed on learning discriminative representations, improving feature alignment, reducing redundant information, and enhancing model robustness across different datasets and acquisition settings. The research also incorporates explainability, fairness, and model calibration to assess the reliability and transparency of automated classification systems.
The developed methodologies are evaluated using publicly available benchmark datasets, including HAM10000, ISIC challenge datasets, PAD-UFES-20, HIBA, and MILK10K. Their effectiveness is assessed through standard classification and reliability metrics, including Accuracy, Balanced Accuracy, Precision, Recall, F1-score, ROC-AUC, calibration error, and statistical significance analysis.
Overall Contribution: The research aims to advance reliable and generalizable deep-learning frameworks for automated skin lesion classification by integrating adaptive feature learning, structured representation optimization, and contrastive learning, while considering computational efficiency and responsible AI principles.
M.E. Thesis Research | July 2011 – June 2012
Title: A Novel Image Steganographic Method Using Octa-Way Pixel-Value Differencing
Thesis Supervisor: Dr. Avinash Gulve, Associate Professor, Department of MCA, Government College of Engineering, Aurangabad, India.
Research Areas: Information Security, Image Steganography, Pixel-Value Differencing, Digital Image Processing, MATLAB.
Research Objective: To develop a secure and imperceptible image steganographic technique for embedding confidential information into digital images while minimizing visual distortion and preserving the perceptual quality of the original image.
Research Description: The M.E. thesis proposes a novel image steganographic technique based on an Octa-Way Pixel-Value Differencing (OPVD) approach for secure information hiding. The method considers each central pixel along with its eight neighboring pixels and computes intensity differences across the corresponding pixel pairs. These differences are utilized to determine the embedding capacity and identify suitable locations for concealing secret information.
Before embedding, the secret information is converted into a binary stream, from which selected bits are embedded into the cover image using the proposed pixel-value differencing strategy. The approach aims to achieve a balance between embedding capacity and imperceptibility while minimizing modifications to the original image.
The performance of the proposed method is evaluated through image-quality and statistical measures, including comparisons of the original and stego-image histograms. The experimental analysis focuses on assessing information-hiding capacity, visual imperceptibility, image distortion, and the effectiveness of the proposed steganographic technique.
Overall Contribution: The work introduces an octa-way pixel-value differencing strategy for image steganography, exploring the use of local pixel relationships to improve secret-data embedding while maintaining the visual quality of the stego image.
B.Tech. Major Project | December 2008 – April 2009
Title: Sachet: An Anomaly-Based Network Intrusion Detection System
Project Supervisor: Dr. U. V. Kulkarni, Professor, Department of Computer Science and Engineering, SGGSIE&T, Nanded, India.
Technologies and Research Areas: Java, JPCAP, MATLAB, Artificial Neural Networks, Network Security, Anomaly Detection, KDD Cup 1999, DARPA.
Project Objective: To develop an anomaly-based network intrusion detection system capable of identifying suspicious network activities through real-time packet analysis and neural-network-based classification.
Project Description: The project, titled Sachet, focuses on developing an anomaly-based Network Intrusion Detection System (NIDS) using artificial neural networks. The proposed system captures network traffic in real time using packet-sniffing techniques and analyzes packet-level information to identify abnormal network activities that may indicate potential security threats.
A neural-network-based detection model is trained using a subset of the KDD Cup 1999 intrusion detection dataset to learn patterns associated with normal and malicious network activities. The trained model is subsequently integrated into the packet-monitoring framework to classify observed network traffic and identify potential intrusions. When suspicious activities are detected, the system generates alerts to notify network administrators.
The project investigates the detection of different categories of network attacks, including active and passive attacks and denial-of-service (DoS) attacks, such as the Smurf attack. The implementation combines Java, JPCAP, and MATLAB to support network packet capture, traffic analysis, and neural-network-based anomaly detection.
Overall Contribution: The project demonstrates the application of artificial neural networks and real-time packet analysis for anomaly-based network intrusion detection, providing an integrated approach to identifying potentially malicious network activities and strengthening network security.
Languages: MATLAB, Python, C, Shell Script
Packages: PyTorch, Tensorflow, Keras, Scikit-Learn, Numpy, LATEX
Selected Subjects: Data Structures, Algorithms, Operating Systems, Parallel Computing, Digital Image Processing, Computer Vision, Machine Learning, Deep Learning, Algorithmic Graph Theory, Bandit Algorithms, Real Time Data Streaming Systems.