Document image analysis

Developed a reasoning-based neural model for information extraction from images of statistical graphs in a question-answering framework with a goal of automatic summarization for the visually impaired (Paper). Designed a connectionist architecture combining deep convolutional and sequence models for end-to-end recognition of offline handwritten text (Paper). Developed a novel framework for automatic digitization of inspection sheets that contain schematics of factory floor, handmarked by engineers. It intelligently combines various traditional image processing methods with an ensemble of deep neural networks to achieve the goal (Paper).