Research Projects


  1. Automatic Karyotyping of Human chromosomes: The aim of this project was to automate the process of karyotyping i.e., segmentation of chromosomes from Q-banded prometaphase images and classification of the segmented chromosomes. The process of karyotyping is usually carried out manually by doctors for diagnosis of birth defects or various disorders. Therefore, we developed various models to automate the process of segmentation and classification in order to assist doctors and reduce their cognitive load.

  2. Extracting information from Document Images: We developed a multi-stage pipeline to automate the process of extracting text regions from document images. It can be utilized in manufacturing units to inspect the health of machines and fill the data in respective log system.

Curricular Projects

Projects completed for the partial fulfillment of my Advanced Deep Learning coursework offered by VideoKen, during my stay at TCS Innovation Labs:

  1. Expedia Hotel Recommendation:
  2. Face Keypoint Detector:
  3. Forecasting Challenge:
  4. Quora Question Pair:
  5. Activity Recognition from videos:
    • Model implemented: CNN + LSTM
    • Tools used: Python, Keras

Projects as a part of my graduate cirruculum:

  1. Shadow Removal and automatic segmentation in OCT images of Optic nerve head: Segmentation of human optic nerve head (ONH) in Optical Coherence Tomography (OCT) images becomes difficult when shadow appear due to presence of blood vessels. We aim to improve the quality of OCT images by removal of shadows and then automatically segmenting the retinal layer of interest in Spectral domain OCT using graph theory and dynamic programming. After extraction of choroid layer using segmentation and SSIM techniques, we focused our interest to extract the blood vessels present in it.
  2. Tools used: MATLAB

  3. Object detection using Support Vector Machines: Binary SVM classifier was trained for car detection with foreground (containing object) and background (without object) images using UIUC image dataset. HOG features were extracted from images to train this model and Non-Maximum Suppression is applied to get boundary box for detected object.
  4. Tools used: MATLAB