About
I’m leading initiatives in Amazon Managed Compute for building Amazon.com’s next…
Experience
Education
-
Indian Institute of Technology (Indian School of Mines), Dhanbad
-
Activities and Societies: CSES Computer Society
Course work and research thesis in Computer Vision & Pattern Recognition with publications in Tier 1 international journals and conferences.
-
-
Electronics & Communication Engineering
-
-
Publications
-
Enhancement of Morphological Snake based Segmentation by imparting Image attachment through Scale-space Continuity
Pattern Recognition Journal / Elsevier
In this paper, we propose a new multi-scale morphological approach to curve evolution useful for object extraction through segmentation. The homogenous image structures that characterize the segmentation process are edges and terminations. Normally the conventional morphological snake (MS) technique employs morphological binary level-set operators for realizing forces. These operations handle definite components of the PDE (partial differential equation) used for modeling the dynamic system…
In this paper, we propose a new multi-scale morphological approach to curve evolution useful for object extraction through segmentation. The homogenous image structures that characterize the segmentation process are edges and terminations. Normally the conventional morphological snake (MS) technique employs morphological binary level-set operators for realizing forces. These operations handle definite components of the PDE (partial differential equation) used for modeling the dynamic system. The proposed model can segment with reasonably high level of accuracy and efficiency while ensuring smooth segmentation at object boundaries with scale space continuity . Application of discrete image force in MS is a per pixel decision based on the sign of image force PDE component. In the continuous domain however, the intensity of the image force PDE component is the primary factor for snake evolution. In our model we embed scale-space continuity into the morphological operators dictated by MS in order to realize the image force both in intensity and direction. Thus, our model confirms to the speed, agility and robustness of morphological snakes with regard to segmentation while ensuring enhanced efficiency of segmentation under noise. We have rated the performance both on qualitative and quantitative basis against benchmark results, on a set of 2D gray-scale real images both in absence and presence of noise. A comparative study has also been carried among our method, MS, geodesic active contour (GAC) and Distance Regularized Level Set Evolution (DRLSE).
Citations - https://scholar.google.com/scholar?oi=bibs&hl=en&cites=290403571034650904Other authors -
Feature binding technique for integration of biological databases with optimized search and retrieve
Procedia Technology/ Elsevier
Biological databases are highly decentralized, having a high degree of difference in terminologies, feature fields, data representation and query formats. This is coupled by the problem of performing multi-database queries manually. Requirement arises therefore to automate the integration of biological databases that do much more than just retrieve and modify data. Speeding up the discovery of new medications and the introduction of new drugs in the market are some additional expectations out…
Biological databases are highly decentralized, having a high degree of difference in terminologies, feature fields, data representation and query formats. This is coupled by the problem of performing multi-database queries manually. Requirement arises therefore to automate the integration of biological databases that do much more than just retrieve and modify data. Speeding up the discovery of new medications and the introduction of new drugs in the market are some additional expectations out of such automation. Feature fields of different biological databases have different formats. To bind a meta-feature to the different feature formats under the same integration platform matching qualifiers is required for the different features. Integration requires binding formats with different databases concurrently, but the high dimensionality and redundancy of the qualifiers makes such integration impossible. Evolutionary selection algorithms have already been applied to reduce high dimensionality in microarray gene expression patterns. Given the similar qualifier redundancy and high qualifier dimensionality for biological databases such as EMBL, GENBANK and DDBJ, multi objective Genetic Algorithm applied to find qualifier reducts is not a misnomer. In feature binding initially Rough set theory is applied to find the initial population of qualifier reduct. Multi Objective Genetic Algorithm (NSGA-II) is run over this population to obtain the exact qualifier reduct. A feature set is categorized with the help of this qualifier reduct. Having done that, the problem of retrieving or manipulating data from a decentralized biological database is addressed in the Search & Retrieve algorithm, where stochastic and machine learning techniques have been used to find high probable warehouses where the data is indexed.
Citations - https://scholar.google.com/scholar?oi=bibs&hl=en&cites=4142758331388613674Other authors -
Concept of Stochastic Memory & Data Retrieval using Artificial Neural Networks
IEEE International Conference on Recent Advances in Information Technology (RAIT 2012)
This paper presents the concept of a physical memory whose state is dependent on a stochastic variable. The stochastic parameter used is temperature. This gives way to efficient space utilization by overlapping data patches upon existing data and overcoming the upper limit of storage space, i.e. more storage data with less hardware and more data security. Furthermore, the paper goes on to present retrieval solutions, for such overlapped data patch structures, using Deep Belief Networks made up…
This paper presents the concept of a physical memory whose state is dependent on a stochastic variable. The stochastic parameter used is temperature. This gives way to efficient space utilization by overlapping data patches upon existing data and overcoming the upper limit of storage space, i.e. more storage data with less hardware and more data security. Furthermore, the paper goes on to present retrieval solutions, for such overlapped data patch structures, using Deep Belief Networks made up of layers of. Restricted Boltzmann machines (RBM), along with mapping with a Bidirectional Associative Memory (BAM).
Projects
-
An Active Surface Model for Segmentation of 3D images using Binary Level-Sets.
Dissertation:
In this project, a novel approach to evolve a non-parametric deformable surface was proposed to 3D objects in volumetric images. A robust edge surface kernel computation technique was proposed with the use of statistical edginess parameters along with application of morphological operator derivatives for evolving a volumetric surface over the aforesaid edge surface kernel space in order to extract 3D volumetric objects under severe noise. Comparative study was carried out with…Dissertation:
In this project, a novel approach to evolve a non-parametric deformable surface was proposed to 3D objects in volumetric images. A robust edge surface kernel computation technique was proposed with the use of statistical edginess parameters along with application of morphological operator derivatives for evolving a volumetric surface over the aforesaid edge surface kernel space in order to extract 3D volumetric objects under severe noise. Comparative study was carried out with the 3D extension of Chan & Vese technique (active contour without edges) and Geodesic Active Surfaces.
-
Segmentation of 2D gray-scale images using Adaptive Morphological Snakes
Minor:
In this project, Partial Differential Equation (PDE) based non-parametric curve evolution technique was applied to track and extract objects. An adaptive morphological derivative of active contour/ snake, was devised for the extraction task. Here, morphological operators were applied to approximate the PDE for curve evolution under scale-space continuity. High order of extraction efficiency was obtained over noisy images with this method in comparison with prevalent Geodesic Active…Minor:
In this project, Partial Differential Equation (PDE) based non-parametric curve evolution technique was applied to track and extract objects. An adaptive morphological derivative of active contour/ snake, was devised for the extraction task. Here, morphological operators were applied to approximate the PDE for curve evolution under scale-space continuity. High order of extraction efficiency was obtained over noisy images with this method in comparison with prevalent Geodesic Active Contour (GAC) and morphological snake (MS) techniques. -
A fast Universal Asynchronous Receiver Transmitter design
Developed a fast UART Tx & Rx system using pipelining was designed and implemented on FPGA kit
using VHDL. -
Hierarchical Design of a 32-bit ALU
Summer Project:
Design and implementation of a 32-bit ALU on XILINX FPGA using VHDL using a hierarchical
and modular Approach.
Honors & Awards
-
Amazon Fulfillment Technology WW - Root cause annihilator
Amazon
Award for excellent deep dive and fixing a long running technical glitch that affected system scalability
-
Amazon Fulfillment Technology WW - Defect Destroyer
Amazon
Award for going the extra mile to check system defects.
-
Amazon AFT India Awards 2016
Amazon
Guardian of Simplicity - Simple solutions to complex technical problems
-
Spot Award
Samsung Research India
Secured 9th rank in Samsung Global SOTONG Programming contest.
Earned TopCoder Badge. -
Employee of the Month
Samsung Research India
Ownership of Samsung Base Email App enhancement frameworks
Languages
-
English
Full professional proficiency
-
Bengali
Native or bilingual proficiency
-
Hindi
Limited working proficiency
Other similar profiles
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore MoreOthers named Sawrav Roy
1 other named Sawrav Roy is on LinkedIn
See others named Sawrav Roy