Self-motivated Data Scientist offering several years of leadership experience across various industries. Methodical with significant experience in data mining and statistical analysis. Excellent problem-solver with history automating processes and driving operational enhancements.
DETECTION OF TEXT IN VIDEO USING DYNAMIC BOUNDING BOX GENERATION WITH FASTER ADAPTIVE REGIONAL PROPOSAL NETWORKS, Researcher Sep 2016 - Jan 2024 Developed an innovative approach for text detection in videos, a critical step in Optical Character Recognition (OCR) systems, with applications in FASTag, document analysis, and smart vehicle navigation. Addressed challenges such as complex backgrounds, poor illumination, and text distortion through hybrid preprocessing techniques. Introduced enhanced text filling, angular rotational gradient, and enhanced Gaussian function to improve edge information and visibility of text data. Implemented a novel feature extraction method using enhanced thinning to extract characters and regions, enabling dynamic bounding box ratio computation. Designed a hybrid architecture incorporating Faster Adaptive Regional Proposal Networks (FARPN) for end-to-end trainable text identification, surpassing traditional methods. Conducted comparative analysis with Faster Region Convolutional Neural Network with a Regional Proposal Network (FRCNN-RPN), demonstrating the superiority of the proposed approach in terms of accuracy and speed. Developed an interface for visualizing detected text in videos. Achieved 91.2% accuracy on the ICDAR 2013-15 "text in video" dataset, showcasing the effectiveness of the FARPN methodology.
Gold Medal in Master of Technology, Best Project in Master of Technology