Department of Computer Science and Engineering,
Indian Institute of Technology Roorkee, 247667, India.
Research Areas: Machine Learning, Image and Video Processing, Medical Imaging, Computer Vision, Activity Recognition, Pattern Recognition, Privacy-Preserving Computing, Data Security.
Data Deduplication over Cloud, Biometric Security.
Dynamic Texture Recognition.
VLSI Architecture for Visibility Enhancement.
VLSI Architecture based high Speed Image Processing.
Emotion AI (Affective Computing).
Data deduplication or intelligent compression is a process that eliminates redundant copies of data and reduces storage overhead. Data deduplication techniques ensure that only one unique instance of data is retained on storage media, such as disk, flash or tape. Redundant data blocks are replaced with a pointer to the unique data copy. In that way, data deduplication closely aligns with incremental backup, which copies only the data that has changed since the previous backup.
Glaciers present in the mountain regions constitute a major part of the earth system. It has been observed that mapping and analyzing glacier components provides us with information which is very important in acting as an input parameter for studies related to glacier mass balance, hydrological and climatological modeling and studies related to glacier hazard. Glacier health could be monitored precisely only by a very accurate temporal analysis of the peri-glacial debris (PGD) and supra-glacial debris (SGD) cover.
Videos are omnipresent in every area from education, entertainment, surveillance, etc. Having large amount of video data imposes restriction on the time required for consumption. Video summarization is the process of generating an abridged version of the given video which encompasses the summary of the video and is of shorter duration.
Human activity recognition is the automatic recognition of human activities in videos. It has wide variety of applications in security surveillance, healthcare, sports and human computer interaction. Hence, there is a need to develop automatic recognition systems using computer vision and machine learning based techniques. Theses algorithm are based on features extracted using handcrafted or state-of-the-art deep learning based techniques. The input modalities can be either RGB, depth or IR videos, while the output is class/activity label.
Over the past couple of years, the field of human activity recognition has made significant progress.The credit of which goes largely to the appearance of bigger datasets (Activity Net etc.) and Deep Learning. The applications of activity recognition are huge, to name a few – video summarization, human behavior analysis, navigation and environmental reconnaissance, elderly health care etc. However, the problem of monocular camera based activity recognition is far from solved, and it is specially challenging to provide unsupervised models for activity recognition. Also, the availability of large number of charge-coupled cameras and unlabeled videos motivates the field of unsupervised activity recognition.
Dynamic texture (DT) is an extension of texture to the temporal domain. DT has two types of information: texture and dynamism. that why in DT, recognition involves analyzing the spatio-temporal domain instead of analyzing each frame separately. Therefore feature descriptors should have the caliber to combine robustness, descriptive power and computational efficiency along with motion for efficiently recognizing
Cloud computing provides cost effective pay-as-you-go services with business continuity. Users are relieved from infrastructure maintenance responsibilities and can accomplish the services directly offered by cloud service providers at reasonable costs. Consequently, new cloud enabled gadgets are evolving nowadays. Regardless of these benefits, users are always concerned about moving their data to the cloud. Hence, new techniques are required to process user's personal images/videos/data without revealing any information and falls in the area of Privacy Preserving Cloud Computing/Encrypted Domain Processing.
Cloud computing is an economically and computationally efficient Information Technology paradigm to provide online data processing paid services. The data can be user's personal images/videos or other confidential documents which user wants to process without revealing any information to the cloud service provider. This user's concern of data privacy is addressed by techniques of Privacy Preserving Cloud Computing in which all data processing algorithms are performed over Encrypted form of data.
Medical Image Processing is an emerging field and plays a key role in segmentation and classifying region of interest (ROI) from bio-medical images. Medical image processing in itself is a broad area of research which has matured using varous methods of diagnosis/screening which could contain x-rays, computer tomography (CT), positron emission tomography(PET), ultrasound (making use of sound waves), magnetic resonance imgaing (MRI) and much more. With the developement of these softwares, detecting and anlysing disesase to their very basic level has now bwwn made possible. This has helped in clear classification and segmentation of any disease.
Image segmentation and classification plays an important role in many applications of medical-imaging by automating or facilitating the delineation of the regions of interest and their classification. Our aim is to provide a semi-automated or automated system to delineate and classify the brain stroke lesion from CT/MR scan images. Brain strokes are mainly of two types- hemorrhagic and ischemic. Hemorrhagic stroke occurs when a weakened blood vessel ruptures and iscehmic stroke occurs when there is an interruption of blood supply to the brain. Therefore, for their treatment accurate segmentation and classification system should exist.
Emotion AI or Affective Computing is an interdisciplinary area combining Cognitive Science and Artificial Intelligence. It deals with gathering the data from voices, faces and body language to identify, understand, process and replicate human affects. The aim is to develop computational systems that could respond according to an end user's state of mind, as well as to understand more about human mind. Technologies like microphones, cameras, sensors, motion trackers, EEG, machine learning, deep learning, optimization, big data, etc. are being utilized in this area.
Biometrics is the study of anatomical or behavioral features of living beings for the purpose of their automatic identification. A generic biometric system has numerous vulnerabilities and security issues associated with it. From intrinsic failures to identity theft, biometric systems can be subjected to a wide variety of adversarial attacks. As a countermeasure, Biometric Template Protection (BTP) schemes ensure the required guarantees of privacy and security to the end users. The aim of this research problem is to design effective BTP schemes which simultaneously provide robust guarantees of minimum information leakage, unlinkability, revocability, irreversibility and performance.
Weather degradation such as haze, fog, mist, etc. severely reduces the effective range of visual surveillance. This degradation is a spatially varying phenomena, which makes this problem non trivial. Weather effects mitigation is an essential preprocessing stage in applications such as long range imaging, border security, intelligent transportation system, etc.