Welcome to my homepage !!!

I am Dr. Vishesh Kumar Tanwar, a postdoctoral scientist currently collaborating with Prof. Sajal Das in the CReWMaN Research Lab at Missouri University of Science and Technology, US. Previously, I had the privilege of working with Prof. Mubarak Shah at the Center for Research in Computer Vision, University of Central Florida, US.

I earned my Ph.D. in Applied Mathematics, where I studied the privacy-preserving algorithms for data processing and statistical analysis techniques using deep learning and obfuscation under the guidance of Prof. Rama Bhargava and Prof. Balasubramanian Raman at the Machine Vision Lab, Indian Institute of Technology Roorkee, India.

Currently, I have been dedicated to crafting privacy-preserving frameworks, focusing on training deep neural networks and statistical analysis using obfuscated data for practical applications in quantitative analysis, temporal data, video surveillance, and smart agriculture.

Research Interests

Machine Learning and Data Processing
Privacy-Preserving Statistical Analysis
Deep Learning
Split Learning
Differential Privacy
Homomorphic Encryption

Google Scholar ResearchGate LinkedIn

Skills

Programming

Python, PyTorch, MATLAB, Scikit-learn, Numpy, Keras, OpenCV; Jupyter, LATEX, and Microsoft Office

Statistical Analysis

Optimization Techniques, Model Scalability, Adversarial Machine Learning, Linear Algebra, Calculus, Probability and Statistics, Algebra.

Machine Learning & Data Science

Applied Computer Vision, Ensemble Learning, Neural

Networks, Predictive Modeling, Deep Learning.

Privacy-preserving Computing

Data Anonymization, Differential Privacy, Federated and Split Learning, Secure Multi-party Computations, Homomorphic Encryption

Some Recent Publications

Figure: An overview of Smart Connected Farms with the tasks executed at different components

Analyzing Real-time Insect Detection in Smart Connected Farms

In agriculture, insects pose a major threat to the quality and quantity of crop yield; consequently, their timely detection is paramount. With a vision of Smart Connected Farms (SCF), this research proposes an Insect Detection framework (InsDet) to identify the most harmful corn crop insect, corn rootworm beetle. InsDet employs an object detection model with varying sizes to localize the insects in sticky trap images. Under the SCF, we collected real-world data from 20 farms in Iowa state, USA, in 2021 and 2022 to train and test our model. We developed a novel prototype (sensor unit), incorporating a resource-constraint device (Raspberry Pi), camera, Long Range (LoRa) module, and wooden sticky trap sheet holder to automatically transfer images to an in-field computing unit (each connected with another via LoRa) periodically and perform real-time inference which takes 1.2 seconds per image to detect target insect.

Figure: Our proposed architecture utilizes an LSTM to generate an expected signal using the previously generated batch to predict the currently expected signal batch.

RaFID: A Lightweight Approach to Radio Frequency Interference Detection in Time Domain using LSTM and Statistical Analysis

Recently, the utilization of Radio Frequency (RF) devices has increased exponentially over numerous vertical platforms. This rise has led to an abundance of Radio Frequency Interference (RFI), which continues to plague RF systems today. The continued crowding of the RF spectrum makes RFI efficient and lightweight mitigation critical. Detecting and localizing the interfering signals is the foremost step for mitigating RFI concerns. Addressing these challenges, we propose a novel and lightweight approach, RaFID, to detect and locate the RFI by incorporating deep neural networks (DNNs) and statistical analysis via batch-wise mean aggregation and standard deviation (SD) calculations. RaFID investigates the generation of an expected signal using DNNs within the time domain. We performed the statistical analysis to compare our generated expected signal with the received signal to detect the existence of interference and determine interference frequency. Experimental results show that signal estimation is accurate, with a mean squared error of 0.012 and an average run-time of 0.5 seconds.

Figure: Comparison of our proposed obfuscated images with varying block sizes. A larger block size offers better privacy.

Preserving Privacy in Image Database through Bit-planes Obfuscation

The recent surge in computer vision applications has caused visual privacy concerns to people who are either users or exposed to an underlying surveillance system. To preserve their privacy, image obfuscation lays out a strong road through which the usability of images can also be maintained without revealing any visual private information. However, prior solutions are susceptible to reconstruction attacks or produce non-trainable images even by leveraging the obfuscation ways. This paper proposes a novel bit-planes-based image obfuscation scheme, called Bimof, to protect the visual privacy of the user in the images that are input into a recognition-based system. By incorporating the chaotic system for non-invertible noise with matrix decomposition, Bimof offers strong security and usability for creating a secure image database. In Bimof, it is hard for an adversary to recover the original image, withstanding a malicious server. We conduct experiments on two standard activity recognition datasets, UCF101 and HMDB51, to validate the effectiveness and usability of our scheme. We provide a rigorous quantitative security analysis through pixel frequency attacks and differential analysis to support our findings.

Figure: The pictorial representation of the Crypt-OR framework. The red-colored star structure is considered the region-to-be-inpaint at the user-end, followed by generating 𝑛 Shamir’s secret shares and transmitted to 𝑛 cloud service providers (CSPs) along with the binary mask 𝑀 indicating the region 𝛺. Each of 𝑛th CSP searches the best-exemplar encrypted patch in their received share 𝑆𝑛 respective of the given mask 𝑀 and then transmits it to the user-end. The user reconstructs each of 𝑛 patches in the plain domain and chooses the most similar patch using the structural similarity metric.

Crypt-OR: A privacy-preserving distributed cloud computing framework for object-removal in the encrypted images

Image inpainting is a technique to modify an image by removing/filling up the undesired region(s) in a visually plausible manner. With the advancement of cloud applications, cloud service providers (CSPs) provide image modifying services such as inpainting and undesired object removal to their users. However, these inpainting services require the user’s original image to contain sensitive information that an adversary can misuse. In this paper, we address these privacy concerns in image enhancement techniques by proposing a novel privacy-preserving distributed-cloud-based framework for object removal in encrypted images, Crypt-OR. The unknown pixel intensity value, obtained by removing the specified object(S), is approximated by incorporating the merits of the exemplar-based search–copy–paste approach and diffusion equation on Shamir’s secret shares. The qualitative and quantitative analysis of Crypt-OR is evaluated over different real-world images with varying shaped and sized objects in the encrypted domain (ED). Crypt-OR outperforms the traditional exemplar-based object-removal schemes and is comparable with generative network-based inpainting schemes in the plain domain (PD). Further, Crypt-OR is proven information-theoretically secure in probabilistic and entropy viewpoints with standard cryptographic adversary attacks.

Figure: A pictorial representation of SecureDL. Two entities user and cloud are involved. The user obfuscates private data D and transmits it to the cloud server. The cloud server owns high-configuration resources and a deep learning model that is trained over the user's obfuscated data.

SecureDL: A privacy preserving deep learning model for image recognition over cloud

The key benefits of cloud services, such as low cost, access flexibility, and mobility, have attracted worldwide users to utilize deep learning algorithms for computer vision. Third parties maintain these cloud servers, and users are always concerned about sharing their confidential data. This paper addressed these concerns by developing SecureDL, a privacy-preserving image recognition model for encrypted data over the cloud. The proposed block-based image encryption scheme is well-designed to protect the image’s visual information. The scheme constitutes an order-preserving permutation ordered binary number system and pseudo-random matrices. The proposed method is proven secure in a probabilistic viewpoint and uses various cryptographic attacks. Experiments are conducted over several image recognition datasets, and the trade-off analytics between the achieved recognition accuracy and data encryption are well described. SecureDL overcomes the storage and computational overheads with fully-homomorphic and multi-party computation-based secure recognition schemes.

Figure: Architecture of CryptoLesion.

CryptoLesion: A Privacy-preserving Model for Lesion Segmentation Using Whale Optimization over Cloud

The low cost, access flexibility, agility, and mobility of cloud infrastructures have attracted medical organizations to store their high-resolution data in encrypted form. Besides storage, these infrastructures provide various image processing services for plain (non-encrypted) images. Meanwhile, the privacy and security of uploaded data depend upon the reliability of the service provider(s). The enforcement of laws towards privacy policies in healthcare organizations, for not disclosing their patient’s sensitive and private medical information, restrict them from utilizing these services. To address these privacy concerns for melanoma detection, we propose CryptoLesion, a privacy-preserving model for segmenting lesion regions using a whale optimization algorithm (WOA) over the cloud in the encrypted domain (ED). The user’s image is encrypted using a permutation-ordered binary number system and a random stumble matrix. The segmentation task is accomplished by dividing an encrypted image into a pre-defined number of clusters whose optimal centroids are obtained by WOA in ED, then assigning each pixel of an encrypted image to the unique centroid. The qualitative and quantitative analysis of CryptoLesion is evaluated over publicly available datasets provided in The International Skin Imaging Collaboration Challenges in 2016, 2017, 2018, and PH2 datasets. The segmented results obtained by CryptoLesion are comparable with the winners of respective challenges. CryptoLesion is secure from a probabilistic viewpoint and various cryptographic attacks. To the best of our knowledge, CryptoLesion is first moving towards the direction of lesion segmentation in ED.

Page updated

Report abuse