Computer Vision, Medical Image Analysis, Machine Learning, Deep Learning & Parallel Computing.
Current Research Activities
Ph.D. Thesis Work Dec 2021 – Present
Title: Analysis of Lesion Images for the Diagnosis of Skin Diseases
Keywords: Medical Image Analysis, Skin Lesion Classification, Deep Learning, Keras, PyTorch
Problem Statement: Development of efficient algorithms for Skin Lesion Classification that are suitable for deployment in real-time & resource-constrained environments.
As part of my Ph.D. research, I am developing innovative and efficient methods for skin disease diagnosis through the analysis of skin lesion images. To accomplish this, I utilize standard datasets such as HAM10000, MedNode, Derm7Pt, PH2, HIBA, PAD-UFES-20, and ISIC challenge datasets (ISIC 2016–2024). The proposed methods are designed to be deployable in real-time scenarios and resource-constrained environments. A major aspect of my research involves tackling challenges like dataset imbalance, variations in lesion shape, size, color, and texture, as well as eliminating unwanted artifacts. To address these issues, I am leveraging and refining advanced techniques, including Attention Mechanisms, Semantic Segmentation, Lightweight CNN Architectures, Transfer Learning, Metric Learning, Meta-Learning, Continual Learning, Domain Adaptation, Disentangled Representation Learning, and Responsible AI. The performance of the developed solutions is generally assessed using standard evaluation metrics such as Balanced Accuracy, Recall, F1-Score, AUC-ROC, IoU, Dice Score, etc.
M.E. Thesis Work July 2011 – June 2012
Title: A Novel Image Steganographic Method Using Octa-Way Pixel-Value Differencing
Thesis Supervisor: Dr. Avinash Gulve, Associate Professor, MCA, GEC, Aurangabad, India
Keywords: Information Security, Image Steganography, Pixel Value Differencing, Matlab
Problem Statement: Development of a secure, imperceptible & robust image steganographic method to hide secret data/information in images.
The proposed solution introduces a novel image steganographic method that employs an octa-way pixel value differencing approach to conceal secret data within an image. Each image pixel is paired with its eight surrounding pixels (octa-way), and the intensity differences across eight such pairs are computed. The secret data is first converted into a binary string, from which specific bits are selected for embedding. The method ensures minimal distortion to the original image, as verified by comparing the histograms of the original and embedded images. By evaluating the embedded image using various performance metrics, the proposed approach demonstrates robustness, imperceptibility, and accuracy in secure data hiding.
B.Tech. Major Project Dec 2008 – Apr 2009
Title: Sachet: An Anomaly Based Network Intrusion Detection System
Keywords: Java, JPCAP, Matlab, Neural Networks, DARPA
Problem Statement: Development of a secure & robust anomaly-based Network Intrusion Detection System.
The proposed solution utilizes a neural network-based approach to detect anomalies in real-time packet data captured through packet sniffers. Identified anomalous packets are labeled as potential threats, indicating possible network intrusions by attackers. The system analyzes packet data to detect and prevent such intrusions, alerting administrators to abnormal activities and enhancing network security. The neural network is trained on a subset of the KDD Cup 1999 dataset, enabling it to recognize and mitigate various types of network attacks, including active and passive attacks, as well as denial-of-service (DoS) attacks (e.g. smurf attack).
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