Logo

Sayan Mandal

Graduate Research Associate
The Ohio State University

Computer Science and Engineering,

2015 Neil Avenue, Dreese Labs 474.

Columbus, OH 43210



Me

About me

I am a Phd Student at the Ohio State University. I am working at the Topological and Data Analysis group and is advised by Dr. Tamal Dey. My research involves computational topology, computer vision and machine learning. We are working on generating topological signatures to explain and classify heirarchical structures of protien molucules. We have submitted our findings in the 'Internation Systems for molucular biology' (ISCB) conference. Previously, we worked on utilising topological signatures as machine learning features. Our work titled 'Improved Image Classification using topological persistence' won the best paper award at the Vision, Modeling and Visualisation Conference in Bohn, Germany 2017.

My research interestes include Computational Topology, Image Processing, Computer Vision, Graphics, Machine Learning, Data Structure and Algorithms. I have signficant coding experiences in C++, Java, Matlab. Recently I have taken a lot of interest into Python as well and find it very powerful.

I have had previous research experiences in Machine Learning and Image Processing. While working as a Research Associate at the Indian Institute of Technology, Kharagpur, we had developed a method to identify and track the variation of the pelvic bones for patients suffering from Colon Cancer using CBCT images. My thesis towards my Master's degree at the Indian Institute of Engineering Science and Technology included numerous image processing and ML techniques on paper based cartographic documents for it's feature extraction, classification and archival.

I have also worked as a Teaching Assistant at IIEST where I conducted lab sessions and problem solving sessions for Computer Graphics and AI. As a T.A. at IIT I held the same responsibilities with Operating Systems. I had to teach the course on Introduction to Computer Programming In Java at OSU during my tenure as a Teaching Associate.

I am from the city of Kolkata in India. It's a beautiful city and in case you are planning to visit, this might be an interesting read. I also love playing the piano and make sure you check out my youtube channel.

Email: mandal(dot)25(at)buckeyemail.osu.edu

   


Technical Skills

Programming Languages:C, C++, Java, Python, Shell script, Python, HTML & CSS
Tools and IDE used: WEKA, Latex, Sublime, Flex, Bison, IDLE, Netbeans, Eclipse, Adobe Flash
Packages used:OpenCV, CGAL, DCMTK, Boost

Publications

Here are a list of my publications:

1

Improved Image Classification using Topological Persistence


Image classification has been a topic of interest for many years. With the advent of Deep Learning, impressive progress has been made on the task, resulting in quite accurate classification. Our work focuses on improving modern image classification techniques by considering topological features as well. We show that incorporating this information allows our models to improve the accuracy, precision and recall on test data, thus providing evidence that topological signatures can be leveraged for enhancing some of the state-of-the art applications in computer vision.
Vision Modeling and Visualisation Workshop(VMV)
2017
2

Study of variation of pelvis positioning for patients suffering from rectal cancer using daily kilo-Voltage Cone Beam CT images


Methods to convert colour images to binary form are already reported in the literature. However, these methods are inadequate for binary conversion of complex documents such as maps due to large intensity variations in different regions and entangled texts with lines representing borders, rivers, roads and other similar components. This paper proposes a new binary conversion technique, for coloured land map images, by extracting the regions and analysing the hue, saturation spread and within class ‘kurtosis’. This is a region-wise adaptive algorithm which copes up with the sharp changes of the discriminating features on different regions. Here, local regions are selected as clusters having the same hues and saturation. These regions are individually converted to binary form using the spread of their degree of within class kurtosis. The individual regions are finally combined. Our experiments include 446 colour maps from the map image database created for this purpose and made freely available at http://code.google.com/p/lmidb
National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG)
2015
3

Map image binarization and stitching using extraction of regions


Research on document image analysis is actively pursued in the last few decades and services like OCR, vectorization of drawings/graphics and various types of form processing are very common. Handwritten documents, old historical documents and documents captured through camera are now being the subjects of active research. However, another very important type of paper document, namely the map document image processing research suffers due to the inherent complexities of the map document and also for nonavailability of benchmark public data-sets. This paper presents a new data-set, namely, the Land Map Image Database (LMIDb) that consists of a variety of land maps images (446 images at present and growing; scanned at 200/300 dpi in TIF format) and the corresponding ground-truth. Using semiautomatic tools non-text part of the images are deleted and the text-only ground-truth is also kept in the database. This paper also presents a classification strategy for map images using which the maps in the database are automatically classified into Political (Po), Physical (Ph), Resource (R) and Topographic (T) maps. The automatic classification of maps help indexing of the images in LMIDb for archival and easy retrieval of the right maps to get the appropriate geographical information. Classification accuracy is also tested on the proposed
Journal of Theoretical and Applied Computer Science
2015
4

Binarisation of Colour Map Images through Extraction of Regions


Methods to convert colour images to binary form are already reported in the literature. However, these methods are inadequate for binary conversion of complex documents such as maps due to large intensity variations in different regions and entangled texts with lines representing borders, rivers, roads and other similar components. This paper proposes a new binary conversion technique, for coloured land map images, by extracting the regions and analysing the hue, saturation spread and within class ‘kurtosis’. This is a region-wise adaptive algorithm which copes up with the sharp changes of the discriminating features on different regions. Here, local regions are selected as clusters having the same hues and saturation. These regions are individually converted to binary form using the spread of their degree of within class kurtosis. The individual regions are finally combined. Our experiments include 446 colour maps from the map image database created for this purpose and made freely available at http://code.google.com/p/lmidb
Springer International Publishing
2014
5

Land Map Image Dataset: Ground-truth and Classification using Visual and Textural Features


Research on document image analysis is actively pursued in the last few decades and services like OCR, vectorization of drawings/graphics and various types of form processing are very common. Handwritten documents, old historical documents and documents captured through camera are now being the subjects of active research. However, another very important type of paper document, namely the map document image processing research suffers due to the inherent complexities of the map document and also for nonavailability of benchmark public data-sets. This paper presents a new data-set, namely, the Land Map Image Database (LMIDb) that consists of a variety of land maps images (446 images at present and growing; scanned at 200/300 dpi in TIF format) and the corresponding ground-truth. Using semiautomatic tools non-text part of the images are deleted and the text-only ground-truth is also kept in the database. This paper also presents a classification strategy for map images using which the maps in the database are automatically classified into Political (Po), Physical (Ph), Resource (R) and Topographic (T) maps. The automatic classification of maps help indexing of the images in LMIDb for archival and easy retrieval of the right maps to get the appropriate geographical information. Classification accuracy is also tested on the proposed data-set and the result is encouraging.
Image Processing & Communications De Gruyter Open
2014
6

Shape Analysis of the Cells in Peripheral Blood Smear for Disease Diagnosis

Disease caused by any parasite can be diagnosedusing the shape of RBCs and WBCs. Manual shape analysismay be faulty. So there is a need in designing an automatedsystem that can identify the shape. This paper presents anapproach to detect the shape of RBCs and WBCs using thehighest, lowest and mean radius of each type of cell. Herecells are segmented using watershed algorithm. The proposedapproach is tested on a collected dataset of microscopic bloodsample images and experimental results are encouraging
Institute of Engineers (India)

Projects

Here are some of the projects I have been involved in while at OSU. For a complete list of projects undertaken during my undergrad and masters, please visit my linkedin profile. The codes for these projects are publicly available at GitHub.

Project Description
Course Instructor
1. Real time rendering of a scene aiming to move a random Droid under variable scene and lightning. Also includes a project on skyboxing and texture allocation.
Real Time Rendering CSE 5542
Hanwei Shen
2. Build a tokenizer, parser, and interpreter to parse and execute a pseudo-Lisp type language including function calls and recursions
Programming language CSE 6341
Nasko Rountev
3. Curve and mesh Generation and reconstruction. Some of the methods include Catmull Clark, Doo-Sabin and NN-Crust
Geometric Modelling CSE 5543
Tamal Dey
4. Mario's Vision: A rendition of the classic Super Mario game. Only this time instead of using keyboard orgaming console controls, players have to perform real-time actions to make Mario move. Theproject is a real-time object and motion detector that estimates human pose and moves Marioaccordingly in its quest to save the princess. Computer Vision CSE 5524
Jim Davis
5. That's Punderful Given an English sentence, attempts to classify words as puns or not puns, using a variety of features and machine learning techniques Machine Learning
CSE 5523
Mikhael Belkin

Honors and Awards